Elixir API Reference

Elixir API Reference v1.16.0

Functions

create_client()

Create a new LLM client with simple scalar configuration.

This is the primary binding entry-point. All parameters except api_key are optional — omitting them uses the same defaults as ClientConfigBuilder.

Errors:

Returns LiterLlmError if the underlying HTTP client cannot be constructed, or if the resolved provider configuration is invalid.

Signature:

@spec create_client(api_key, base_url, timeout_secs, max_retries, model_hint) :: {:ok, term()} | {:error, term()}
def create_client(api_key, base_url, timeout_secs, max_retries, model_hint)

Example:

{:ok, result} = create_client("value", "value", 42, 42, "value")

Parameters:

Name	Type	Required	Description
`api_key`	`String.t()`	Yes	The api key
`base_url`	`String.t() \| nil`	No	The base url
`timeout_secs`	`integer() \| nil`	No	The timeout secs
`max_retries`	`integer() \| nil`	No	The max retries
`model_hint`	`String.t() \| nil`	No	The model hint

Returns: DefaultClient

Errors: Returns {:error, reason}

create_client_from_json()

Create a new LLM client from a JSON string.

The JSON object accepts the same fields as liter-llm.toml (snake_case).

Errors:

Returns LiterLlmError.BadRequest if json is not valid JSON or contains unknown fields.

Signature:

@spec create_client_from_json(json) :: {:ok, term()} | {:error, term()}
def create_client_from_json(json)

Example:

{:ok, result} = create_client_from_json("value")

Parameters:

Name	Type	Required	Description
`json`	`String.t()`	Yes	The json

Returns: DefaultClient

Errors: Returns {:error, reason}

encode_data_url()

Encode bytes as a base64 data URL: data:<mime>;base64,<b64>.

mime defaults to IMAGE_PNG when nil.

Signature:

@spec encode_data_url(bytes, mime) :: {:ok, term()} | {:error, term()}
def encode_data_url(bytes, mime)

Example:

{:ok, result} = encode_data_url(<<100, 97, 116, 97>>, "value")

Parameters:

Name	Type	Required	Description
`bytes`	`binary()`	Yes	The bytes
`mime`	`String.t() \| nil`	No	The mime

Returns: String.t()

decode_data_url()

Decode a base64 data URL into DecodedDataUrl.

Returns nil for:

Non-data URLs (strings that do not start with "data:").
Malformed prefixes (missing ";base64," marker).
Invalid base64 payloads.

The returned MIME string is extracted verbatim from the URL prefix — it is not validated or normalised.

Signature:

@spec decode_data_url(url) :: {:ok, term()} | {:error, term()}
def decode_data_url(url)

Example:

{:ok, result} = decode_data_url("value")

Parameters:

Name	Type	Required	Description
`url`	`String.t()`	Yes	The URL to fetch

Returns: DecodedDataUrl | nil

register_custom_provider()

The provider will be checked before all built-in providers during model detection. If a provider with the same name already exists it is replaced.

Errors:

Returns an error if the config is invalid (empty name, empty base_url, or no model prefixes).

Signature:

@spec register_custom_provider(config) :: {:ok, term()} | {:error, term()}
def register_custom_provider(config)

Example:

:ok = register_custom_provider(%{{}})

Parameters:

Name	Type	Required	Description
`config`	`CustomProviderConfig`	Yes	The configuration options

Returns: No return value.

Errors: Returns {:error, reason}

unregister_custom_provider()

Remove a previously registered custom provider by name.

Returns true if a provider with the given name was found and removed, false if no such provider existed.

Errors:

Returns an error if the custom-provider registry cannot be updated.

Signature:

@spec unregister_custom_provider(name) :: {:ok, term()} | {:error, term()}
def unregister_custom_provider(name)

Example:

{:ok, result} = unregister_custom_provider("value")

Parameters:

Name	Type	Required	Description
`name`	`String.t()`	Yes	The name

Returns: boolean()

Errors: Returns {:error, reason}

capabilities()

Return the capability flags for a named provider.

Performs an O(n) linear scan over the embedded registry (165 entries). Returns an owned value so bindings can pass capability data without borrowing registry internals.

For unknown provider_name values the function returns an all-false sentinel so callers never need to handle Option.

Signature:

@spec capabilities(provider_name) :: {:ok, term()} | {:error, term()}
def capabilities(provider_name)

Example:

{:ok, result} = capabilities("value")

Parameters:

Name	Type	Required	Description
`provider_name`	`String.t()`	Yes	The provider name

Returns: ProviderCapabilities

all_providers()

Return all provider configs from the registry.

Useful for tooling, documentation generation, or runtime enumeration. Returns the public ProviderConfig slice (without capability flags). To query capability flags for a specific provider use capabilities.

Signature:

@spec all_providers() :: {:ok, term()} | {:error, term()}
def all_providers()

Example:

{:ok, result} = all_providers()

Returns: list(ProviderConfig)

Errors: Returns {:error, reason}

complex_provider_names()

Return the set of complex provider names.

Complex providers require custom auth/routing logic beyond simple bearer tokens (e.g. AWS Bedrock SigV4, Vertex AI OAuth2).

The returned reference points into the static registry — no allocation.

Signature:

@spec complex_provider_names() :: {:ok, term()} | {:error, term()}
def complex_provider_names()

Example:

{:ok, result} = complex_provider_names()

Returns: list(String.t())

Errors: Returns {:error, reason}

completion_cost()

Calculate the estimated cost of a completion given a model name and token counts.

Returns nil if the model is not present in the embedded pricing registry. Returns Some(cost_usd) otherwise, where the value is in US dollars.

When an exact model name match is not found, progressively shorter prefixes are tried by stripping from the last - or . separator. For example, gpt-4-0613 will match gpt-4 if no gpt-4-0613 entry exists.

Signature:

@spec completion_cost(model, prompt_tokens, completion_tokens) :: {:ok, term()} | {:error, term()}
def completion_cost(model, prompt_tokens, completion_tokens)

Example:

{:ok, result} = completion_cost("value", 42, 42)

Parameters:

Name	Type	Required	Description
`model`	`String.t()`	Yes	The model
`prompt_tokens`	`integer()`	Yes	The prompt tokens
`completion_tokens`	`integer()`	Yes	The completion tokens

Returns: float() | nil

completion_cost_with_cache()

Calculate the estimated cost of a completion, accounting for cached (cache-hit) prompt tokens billed at the provider’s discounted rate.

cached_tokens is the count of prompt tokens served from the provider’s prompt cache. It must be <= prompt_tokens (cached tokens are a subset of the prompt). The non-cached portion is billed at input_cost_per_token and the cached portion at cache_read_input_token_cost when the model has cache pricing; otherwise the entire prompt is billed at the regular input rate.

Returns nil if the model is not present in the embedded pricing registry, mirroring completion_cost.

When the model has ModelPricing.tiers, the tier whose min_context_tokens is the highest value <= prompt_tokens supplies the input/output/cache rates for the whole call; models without tiers (or when prompt_tokens is below every tier threshold) use the base rates unchanged, matching the original flat-rate behaviour.

Signature:

@spec completion_cost_with_cache(model, prompt_tokens, cached_tokens, completion_tokens) :: {:ok, term()} | {:error, term()}
def completion_cost_with_cache(model, prompt_tokens, cached_tokens, completion_tokens)

Example:

{:ok, result} = completion_cost_with_cache("value", 42, 42, 42)

Parameters:

Name	Type	Required	Description
`model`	`String.t()`	Yes	The model
`prompt_tokens`	`integer()`	Yes	The prompt tokens
`cached_tokens`	`integer()`	Yes	The cached tokens
`completion_tokens`	`integer()`	Yes	The completion tokens

Returns: float() | nil

model_info()

Look up FFI-friendly pricing and capability metadata for a model.

Returns nil if the model is not present in the active pricing registry. Uses the same exact-match-then-prefix-fallback resolution as model_pricing; unlike model_pricing, the result is an owned ModelInfo value safe to hand across the FFI boundary.

When a runtime catalog refresh has succeeded, this reflects the refreshed (overlay) catalog; otherwise it reflects the embedded catalog. See model_pricing for the embedded-only alternative.

Signature:

@spec model_info(model) :: {:ok, term()} | {:error, term()}
def model_info(model)

Example:

{:ok, result} = model_info("value")

Parameters:

Name	Type	Required	Description
`model`	`String.t()`	Yes	The model

Returns: ModelInfo | nil

clear()

Remove all guardrails from the global registry.

Primarily useful in tests to reset state between test cases.

Panics:

Panics if the global registry lock is poisoned.

Signature:

@spec clear() :: {:ok, term()} | {:error, term()}
def clear()

Example:

:ok = clear()

Returns: No return value.

count_tokens()

Count tokens in a text string using the tokenizer for the given model.

The tokenizer is resolved from the model name prefix (e.g. "gpt-4o" maps to the Xenova/gpt-4o HuggingFace tokenizer). Tokenizers are cached after first load.

Errors:

Returns LiterLlmError.BadRequest if the tokenizer cannot be loaded (e.g. network failure on first use) or if tokenization itself fails.

Signature:

@spec count_tokens(model, text) :: {:ok, term()} | {:error, term()}
def count_tokens(model, text)

Example:

{:ok, result} = count_tokens("value", "value")

Parameters:

Name	Type	Required	Description
`model`	`String.t()`	Yes	The model
`text`	`String.t()`	Yes	The text

Returns: integer()

Errors: Returns {:error, reason}

count_request_tokens()

Count tokens for a full ChatCompletionRequest.

Sums tokens across all message text contents plus a per-message overhead of ~4 tokens (for role, separators, and formatting metadata). Tool definitions and multimodal content parts (images, audio, documents) are not counted — only textual content contributes to the token total.

Errors:

Returns LiterLlmError.BadRequest if the tokenizer cannot be loaded or if tokenization fails for any message.

Signature:

@spec count_request_tokens(model, req) :: {:ok, term()} | {:error, term()}
def count_request_tokens(model, req)

Example:

{:ok, result} = count_request_tokens("value", %{{}})

Parameters:

Name	Type	Required	Description
`model`	`String.t()`	Yes	The model
`req`	`ChatCompletionRequest`	Yes	The chat completion request

Returns: integer()

Errors: Returns {:error, reason}

check_bound()

Assert that current_len + incoming does not exceed limit.

Call this before appending incoming bytes to any buffer that must stay below limit. Returns Err(LiterLlmError.Streaming) on overflow and emits a tracing.warn! with context.

Signature:

@spec check_bound(context, current_len, incoming, limit) :: {:ok, term()} | {:error, term()}
def check_bound(context, current_len, incoming, limit)

Example:

:ok = check_bound("value", 42, 42, 42)

Parameters:

Name	Type	Required	Description
`context`	`String.t()`	Yes	The context
`current_len`	`integer()`	Yes	The current len
`incoming`	`integer()`	Yes	The incoming
`limit`	`integer()`	Yes	The limit

Returns: No return value.

Errors: Returns {:error, reason}

ensure_crypto_provider()

Install the ring crypto provider as the rustls process default, idempotently.

rustls 0.23+ removed the implicit default provider. This function installs ring once per process. Subsequent calls are no-ops. Calling it after another rustls crypto provider has already been installed is safe: the Err from install_default() is silently ignored.

Called automatically by every internal reqwest.Client constructor (auth providers, default HTTP client). Bindings and downstream consumers reach those constructors transitively, so no manual init is required.

WASM builds are exempt — the WASM target uses the browser/Node.js fetch API instead of rustls, so no crypto provider is needed.

Windows builds use native-tls (SChannel) via reqwest, so rustls is not present and no crypto provider installation is needed.

Signature:

@spec ensure_crypto_provider() :: {:ok, term()} | {:error, term()}
def ensure_crypto_provider()

Example:

:ok = ensure_crypto_provider()

Returns: No return value.

ensure_crypto_provider()

No-op on Windows: reqwest uses native-tls (SChannel), so no rustls provider installation is needed. All callers use the same call site regardless of platform.

Signature:

@spec ensure_crypto_provider() :: {:ok, term()} | {:error, term()}
def ensure_crypto_provider()

Example:

:ok = ensure_crypto_provider()

Returns: No return value.

install_catalog_overlay_from_str()

Install the overlay registry from a raw catalog JSON string, bypassing the network and disk cache entirely.

Parses and flattens catalog_json with the same registry_from_catalog_str logic used for the embedded catalog and the network refresh path, then atomically swaps it in as the active overlay. A parse failure returns CatalogRefreshError.Parse and leaves any existing overlay untouched.

This is primarily a testable seam: it lets tests exercise overlay installation and the embedded/overlay fallback behavior in completion_cost / model_info without a real network call.

Signature:

@spec install_catalog_overlay_from_str(catalog_json) :: {:ok, term()} | {:error, term()}
def install_catalog_overlay_from_str(catalog_json)

Example:

:ok = install_catalog_overlay_from_str("value")

Parameters:

Name	Type	Required	Description
`catalog_json`	`String.t()`	Yes	The catalog json

Returns: No return value.

Errors: Returns {:error, reason}

clear_catalog_overlay()

Clear the overlay registry, reverting completion_cost, completion_cost_with_cache, and model_info to the embedded catalog.

Primarily a test seam (see install_catalog_overlay_from_str); also usable by long-running processes that want to abandon a runtime refresh.

Signature:

@spec clear_catalog_overlay() :: {:ok, term()} | {:error, term()}
def clear_catalog_overlay()

Example:

:ok = clear_catalog_overlay()

Returns: No return value.

refresh_catalog()

Refresh the runtime catalog overlay per config.

config.enabled == false: returns Ok(RefreshOutcome.Disabled) immediately. No network, filesystem, or overlay activity.
A fresh on-disk cache (age < config.ttl_seconds) exists at the resolved cache path (config.cache_path, or a default under std.env.temp_dir()): read + flatten it and install the overlay, returning Ok(RefreshOutcome.FromCache). No network request is made.
Otherwise: validate config.source_url uses https (CatalogRefreshError.InsecureUrl otherwise), fetch it, flatten it, install the overlay, best-effort write the raw JSON to the cache path (a cache write failure does not fail the refresh), and return Ok(RefreshOutcome.Fetched).

On any error return, the overlay is left untouched: the previously active registry (a prior successful overlay, or the embedded catalog if none was ever installed) remains in effect. This is what makes the feature air-gap-safe — an unreachable or invalid source_url never degrades completion_cost / model_info below embedded-catalog availability.

Signature:

@spec refresh_catalog(config) :: {:ok, term()} | {:error, term()}
def refresh_catalog(config)

Example:

{:ok, result} = refresh_catalog(%{{}})

Parameters:

Name	Type	Required	Description
`config`	`CatalogRefreshConfig`	Yes	The configuration options

Returns: RefreshOutcome

Errors: Returns {:error, reason}

Types

AssistantMessage

Assistant’s response to a user message.

Field	Type	Default	Description
`content`	`AssistantContent \| nil`	`nil`	The assistant’s response: plain text, structured parts, or absent. `nil` is valid when the model replies with tool calls only.
`name`	`String.t() \| nil`	`nil`	Optional name for the assistant.
`tool_calls`	`list(ToolCall) \| nil`	`\[\]`	Tool calls the model wants to execute, if any.
`refusal`	`String.t() \| nil`	`nil`	Refusal reason, if the model declined to respond per safety policies.
`function_call`	`FunctionCall \| nil`	`nil`	Deprecated legacy function_call field; retained for API compatibility.
`reasoning_content`	`String.t() \| nil`	`nil`	Reasoning/thinking tokens returned by the provider, if any (e.g. DeepSeek R1, Qwen `reasoning_content`, or Anthropic extended thinking).

Functions

text()

Return the assistant’s textual response, concatenating all Text parts if the content is structured.

Returns nil for Refusal-only or OutputImage-only responses.

Signature:

def text()

Example:

{:ok, result} = instance.text()

Returns: String.t() | nil

refusal_text()

Return the refusal message, if the model declined to respond.

Checks both the top-level refusal field and any Refusal parts inside a structured content.

Signature:

def refusal_text()

Example:

{:ok, result} = instance.refusal_text()

Returns: String.t() | nil

reasoning_text()

Return the model’s reasoning/thinking tokens, if the provider returned any.

Signature:

def reasoning_text()

Example:

{:ok, result} = instance.reasoning_text()

Returns: String.t() | nil

output_images()

Return all AssistantPart.OutputImage parts in the response.

Signature:

def output_images()

Example:

{:ok, result} = instance.output_images()

Returns: list(ImageUrl)

output_audio()

Return all AssistantPart.OutputAudio parts in the response.

Signature:

def output_audio()

Example:

{:ok, result} = instance.output_audio()

Returns: list(AudioContent)

AudioContent

Audio content part for speech-capable models.

Field	Type	Default	Description
`data`	`String.t()`	—	Base64-encoded audio data.
`format`	`String.t()`	—	Audio format (e.g., “wav”, “mp3”, “ogg”).

AuthConfig

Auth configuration block.

Field	Type	Default	Description
`auth_type`	`AuthType`	—	Auth scheme classification.
`env_var`	`String.t() \| nil`	`nil`	Name of the environment variable that holds the API key (e.g. `"OPENAI_API_KEY"`). Holds the variable name, never the secret value.

BatchListQuery

Query parameters for listing batches.

Field	Type	Default	Description
`limit`	`integer() \| nil`	`nil`	Maximum number of results to return. Defaults to 20.
`after`	`String.t() \| nil`	`nil`	Pagination cursor: return results after this batch ID.

BatchListResponse

Response from listing batches.

Field	Type	Default	Description
`object`	`String.t()`	—	Object type (always `"list"`).
`data`	`list(BatchObject)`	`\[\]`	List of batch objects.
`has_more`	`boolean() \| nil`	`nil`	Whether more results are available.
`first_id`	`String.t() \| nil`	`nil`	First batch ID in the result set (for pagination).
`last_id`	`String.t() \| nil`	`nil`	Last batch ID in the result set (for pagination).

BatchObject

A batch job object.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique batch ID.
`object`	`String.t()`	—	Object type (always `"batch"`).
`endpoint`	`String.t()`	—	API endpoint (e.g., `"/v1/chat/completions"`).
`input_file_id`	`String.t()`	—	ID of the input file.
`completion_window`	`String.t()`	—	Completion window (e.g., `"24h"`).
`status`	`BatchStatus`	`:validating`	Current job status.
`output_file_id`	`String.t() \| nil`	`nil`	ID of the output file (present when completed).
`error_file_id`	`String.t() \| nil`	`nil`	ID of the error file (present if some requests failed).
`created_at`	`integer()`	—	Unix timestamp of batch creation.
`completed_at`	`integer() \| nil`	`nil`	Unix timestamp of completion (if completed).
`failed_at`	`integer() \| nil`	`nil`	Unix timestamp of failure (if failed).
`expired_at`	`integer() \| nil`	`nil`	Unix timestamp of expiration (if expired).
`request_counts`	`BatchRequestCounts \| nil`	`nil`	Request processing counts.
`metadata`	`term() \| nil`	`nil`	Metadata attached to the batch.

BatchRequestCounts

Request processing counts for a batch.

Field	Type	Default	Description
`total`	`integer()`	—	Total requests in the batch.
`completed`	`integer()`	—	Completed requests.
`failed`	`integer()`	—	Failed requests.

BedrockConfig

AWS Bedrock configuration.

All fields are optional; anything left unset falls back to the standard AWS environment variables (AWS_DEFAULT_REGION / AWS_REGION, AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN, BEDROCK_CROSS_REGION).

Field	Type	Default	Description
`region`	`String.t() \| nil`	`nil`	AWS region (e.g. `"us-east-1"`).
`cross_region_prefix`	`String.t() \| nil`	`nil`	Cross-region inference profile prefix (e.g. `"us"`).
`access_key_id`	`String.t() \| nil`	`nil`	Explicit AWS access key ID.
`secret_access_key`	`String.t() \| nil`	`nil`	Explicit AWS secret access key.
`session_token`	`String.t() \| nil`	`nil`	Explicit AWS session token (temporary credentials).

BudgetConfig

Configuration for budget enforcement.

Field	Type	Default	Description
`global_limit`	`float() \| nil`	`nil`	Maximum total spend across all models, in USD. `nil` means unlimited.
`model_limits`	`map()`	`%{}`	Per-model spending limits in USD. Models not listed here are only constrained by `global_limit`.
`enforcement`	`Enforcement`	`:hard`	Whether to reject requests or merely warn when a limit is exceeded.

Functions

default()

Signature:

def default()

Example:

{:ok, result} = BudgetConfig.default()

Returns: BudgetConfig

CacheConfig

Configuration for the response cache.

Field	Type	Default	Description
`max_entries`	`integer()`	`256`	Maximum number of cached entries.
`ttl`	`integer()`	`300000ms`	Time-to-live for each cached entry.
`backend`	`CacheBackend`	`:memory`	Storage backend to use.

Functions

default()

Signature:

def default()

Example:

{:ok, result} = CacheConfig.default()

Returns: CacheConfig

CatalogRefreshConfig

Plain-data configuration for refresh_catalog.

Deliberately FFI/binding-friendly: no Duration or PathBuf, just primitives that translate directly across language boundaries.

Field	Type	Default	Description
`enabled`	`boolean()`	`false`	Runtime catalog refresh is entirely opt-in: when `false`, `refresh_catalog` is a no-op that returns `Ok(RefreshOutcome.Disabled)` without touching the network, the filesystem, or the overlay registry.
`source_url`	`String.t()`	—	Source URL to fetch `catalog.json` from. Must be `https`. Defaults to `DEFAULT_CATALOG_URL`; configurable so self-hosted mirrors work.
`ttl_seconds`	`integer()`	`86400`	How long a cached `catalog.json` remains valid before a network refetch is attempted, in seconds.
`cache_path`	`String.t() \| nil`	`nil`	Filesystem path for the on-disk cache. `nil` uses a default path under `std.env.temp_dir()`.

Functions

default()

Signature:

def default()

Example:

{:ok, result} = CatalogRefreshConfig.default()

Returns: CatalogRefreshConfig

ChatCompletionChunk

A streamed chunk of a chat completion response.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique identifier for this stream.
`object`	`String.t()`	—	Always `"chat.completion.chunk"` from OpenAI-compatible APIs. Stored as a plain `String` so non-standard provider values do not fail parsing.
`created`	`integer()`	—	Unix timestamp of chunk creation.
`model`	`String.t()`	—	Model used to generate the chunk.
`choices`	`list(StreamChoice)`	`\[\]`	Streaming choices (delta updates).
`usage`	`Usage \| nil`	`nil`	Token usage (typically only in the final chunk).
`system_fingerprint`	`String.t() \| nil`	`nil`	Fingerprint of the system configuration (OpenAI-specific).
`service_tier`	`String.t() \| nil`	`nil`	Service tier used (OpenAI-specific).

ChatCompletionRequest

Chat completion request (compatible with OpenAI and similar APIs).

Field	Type	Default	Description
`model`	`String.t()`	—	Model ID (e.g., `"gpt-4o-mini"`, `"claude-3-5-sonnet"`).
`messages`	`list(Message)`	`\[\]`	Conversation history from oldest to newest.
`temperature`	`float() \| nil`	`nil`	Sampling temperature in `\[0.0, 2.0\]`. Higher increases randomness. Defaults to 1.0.
`top_p`	`float() \| nil`	`nil`	Nucleus sampling parameter in `\[0.0, 1.0\]`. Lower is more focused.
`n`	`integer() \| nil`	`nil`	Number of chat completions to generate. Defaults to 1.
`stream`	`boolean() \| nil`	`nil`	Whether to stream the response. Managed by the client layer — do not set directly.
`stop`	`StopSequence \| nil`	`nil`	Stop sequence(s) that halt token generation.
`max_tokens`	`integer() \| nil`	`nil`	Max output tokens. Different from max_completion_tokens in some providers.
`presence_penalty`	`float() \| nil`	`nil`	Presence penalty in `\[-2.0, 2.0\]`. Positive discourages repeated topics.
`frequency_penalty`	`float() \| nil`	`nil`	Frequency penalty in `\[-2.0, 2.0\]`. Positive discourages repeated tokens.
`logit_bias`	`map() \| nil`	`%{}`	Token bias map. Uses `BTreeMap` (sorted keys) for deterministic serialization order — important when hashing or signing requests.
`user`	`String.t() \| nil`	`nil`	User identifier for request tracking and abuse detection.
`tools`	`list(ChatCompletionTool) \| nil`	`\[\]`	Tools the model can invoke.
`tool_choice`	`ToolChoice \| nil`	`nil`	Tool usage mode (auto, required, none, or specific tool).
`parallel_tool_calls`	`boolean() \| nil`	`nil`	Whether the model can call multiple tools in parallel. Defaults to true.
`response_format`	`ResponseFormat \| nil`	`nil`	Output format constraint (text, JSON, JSON schema).
`stream_options`	`StreamOptions \| nil`	`nil`	Streaming options (e.g., include_usage).
`seed`	`integer() \| nil`	`nil`	Random seed for reproducible outputs. Provider support varies.
`reasoning_effort`	`ReasoningEffort \| nil`	`nil`	Reasoning effort level (minimal, low, medium, high, max) for extended-thinking models.
`modalities`	`list(Modality) \| nil`	`\[\]`	Output modalities to request from the model. For OpenAI audio models, pass `\["text", "audio"\]`. Vertex AI / Gemini translates these to `generationConfig.responseModalities` (uppercase).
`extra_body`	`term() \| nil`	`nil`	Provider-specific extra parameters merged into the request body. Use for guardrails, safety settings, grounding config, etc.

ChatCompletionResponse

Chat completion response from the API.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique identifier for this response.
`object`	`String.t()`	—	Always `"chat.completion"` from OpenAI-compatible APIs. Stored as a plain `String` so non-standard provider values do not break deserialization.
`created`	`integer()`	—	Unix timestamp of response creation.
`model`	`String.t()`	—	Model used to generate the response.
`choices`	`list(Choice)`	`\[\]`	List of completion choices.
`usage`	`Usage \| nil`	`nil`	Token usage statistics.
`system_fingerprint`	`String.t() \| nil`	`nil`	Fingerprint of the system configuration (OpenAI-specific).
`service_tier`	`String.t() \| nil`	`nil`	Service tier used (OpenAI-specific).

ChatCompletionTool

A tool the model can invoke (currently, all tools are functions).

Field	Type	Default	Description
`tool_type`	`ToolType`	—	Tool type (always “function” in OpenAI spec).
`function`	`FunctionDefinition`	—	Function definition with name, description, and JSON schema parameters.

Choice

A single completion choice.

Field	Type	Default	Description
`index`	`integer()`	—	Index of this choice in the choices array.
`message`	`AssistantMessage`	—	The assistant’s message response.
`finish_reason`	`FinishReason \| nil`	`nil`	Why the model stopped generating (stop, length, tool_calls, content_filter, etc.).

ChunkMiddleware

A per-chunk transformation in the StreamPipeline.

Each middleware receives a typed chunk and returns Ok(Some(chunk)) to pass it through (optionally modified), Ok(None) to drop the chunk, or Err(e) to propagate a stream error.

The trait is object-safe so multiple middleware implementations can be chained inside StreamPipeline.

Functions

process()

Process a single chunk.

Ok(Some(chunk)) — emit (possibly transformed) chunk.
Ok(None) — drop this chunk silently.
Err(e) — propagate as a stream error.

Signature:

def process(chunk)

Example:

{:ok, result} = instance.process(%{{}})

Parameters:

Name	Type	Required	Description
`chunk`	`ChatCompletionChunk`	Yes	The chat completion chunk

Returns: ChatCompletionChunk | nil

Errors: Returns {:error, reason}

CreateBatchRequest

Request to create a batch job.

Field	Type	Default	Description
`input_file_id`	`String.t()`	—	ID of the uploaded input file (JSONL format).
`endpoint`	`String.t()`	—	API endpoint (e.g., `"/v1/chat/completions"`).
`completion_window`	`String.t()`	—	Completion window (e.g., `"24h"`).
`metadata`	`term() \| nil`	`nil`	Optional metadata to attach to the batch.

CreateFileRequest

Request to upload a file.

Field	Type	Default	Description
`file`	`String.t()`	—	Base64-encoded file data.
`purpose`	`FilePurpose`	`:assistants`	Purpose for the file.
`filename`	`String.t() \| nil`	`nil`	Optional filename to associate with the upload.

CreateImageRequest

Request to create images from a text prompt.

Field	Type	Default	Description
`prompt`	`String.t()`	—	Text description of the image to generate.
`model`	`String.t() \| nil`	`nil`	Model ID (e.g., `"dall-e-3"`). Optional; API may use default if unset.
`n`	`integer() \| nil`	`nil`	Number of images to generate. Defaults to 1.
`size`	`String.t() \| nil`	`nil`	Image size (e.g., `"1024x1024"`, `"1792x1024"`).
`quality`	`String.t() \| nil`	`nil`	Image quality: `"standard"` or `"hd"`.
`style`	`String.t() \| nil`	`nil`	Style: `"natural"` or `"vivid"` (DALL-E 3 only).
`response_format`	`String.t() \| nil`	`nil`	Response format: `"url"` or `"b64_json"`.
`user`	`String.t() \| nil`	`nil`	User identifier for request tracking.

CreateResponseRequest

Request to create a structured response.

Field	Type	Default	Description
`model`	`String.t()`	—	Model ID.
`input`	`term()`	—	Input data to process (e.g., a document to extract from).
`instructions`	`String.t() \| nil`	`nil`	Instructions for processing the input.
`tools`	`list(ResponseTool) \| nil`	`\[\]`	Available tools the model can use.
`temperature`	`float() \| nil`	`nil`	Sampling temperature in `\[0.0, 2.0\]`. Defaults to 1.0.
`max_output_tokens`	`integer() \| nil`	`nil`	Maximum output tokens.
`metadata`	`term() \| nil`	`nil`	Optional metadata.
`stream`	`boolean() \| nil`	`nil`	Whether to stream the response. Managed by the client layer — do not set directly.

CreateSpeechRequest

Request to generate speech audio from text.

Field	Type	Default	Description
`model`	`String.t()`	—	Model ID (e.g., `"tts-1"`, `"tts-1-hd"`).
`input`	`String.t()`	—	Text to synthesize into speech.
`voice`	`String.t()`	—	Voice name (e.g., `"alloy"`, `"echo"`, `"fable"`, `"onyx"`, `"nova"`, `"shimmer"`).
`response_format`	`String.t() \| nil`	`nil`	Audio format (e.g., `"mp3"`, `"opus"`, `"aac"`, `"flac"`, `"wav"`, `"pcm"`).
`speed`	`float() \| nil`	`nil`	Playback speed in `\[0.25, 4.0\]`. Defaults to 1.0.

CreateTranscriptionRequest

Request to transcribe audio into text.

Field	Type	Default	Description
`model`	`String.t()`	—	Model ID (e.g., `"whisper-1"`).
`file`	`String.t()`	—	Base64-encoded audio file data.
`language`	`String.t() \| nil`	`nil`	Language ISO-639-1 code (e.g., `"en"`, `"fr"`, `"de"`). Optional; model auto-detects.
`prompt`	`String.t() \| nil`	`nil`	Optional text to guide the model (improves accuracy for domain-specific terms).
`response_format`	`String.t() \| nil`	`nil`	Output format (e.g., `"json"`, `"text"`, `"vtt"`, `"srt"`, `"verbose_json"`).
`temperature`	`float() \| nil`	`nil`	Sampling temperature in `\[0.0, 1.0\]`. Higher increases variability. Defaults to 0.

CustomProviderConfig

Configuration for registering a custom LLM provider at runtime.

Field	Type	Default	Description
`name`	`String.t()`	—	Unique name for this provider (e.g., “my-provider”).
`base_url`	`String.t()`	—	Base URL for the provider’s API (e.g., `<https://api.my-provider.com/v1>`).
`auth_header`	`AuthHeaderFormat`	—	Authentication header format.
`model_prefixes`	`list(String.t())`	—	Model name prefixes that route to this provider (e.g., `\["my-"\]`).

DecodedDataUrl

Result of decoding a data: URL — MIME type and the decoded byte payload.

Named struct (rather than a tuple) so polyglot bindings can extract decode_data_url with a typed return rather than a sanitized scalar.

Field	Type	Default	Description
`mime`	`String.t()`	—	MIME type extracted from the URL prefix (verbatim, not normalised).
`data`	`binary()`	—	Decoded base64 payload.

DefaultClient

Default client implementation backed by reqwest.

Sends requests to 165 LLM providers with automatic provider detection and per-request routing. The provider is resolved at construction time from model_hint (or defaults to OpenAI), but individual requests can override the provider via model name prefix (e.g. "anthropic/claude-3-5-sonnet" routes to Anthropic regardless of construction-time setting).

When the model prefix does not match any known provider, the construction-time provider is used as the fallback. This enables seamless migration between providers by changing only the model name.

The provider is stored behind an Arc so it can be shared cheaply into async closures and streaming tasks. Pre-computed auth headers and extra headers are cached at construction to avoid redundant encoding on every request.

Functions

fetch_batch_for_polling()

Signature:

def fetch_batch_for_polling(batch_id)

Example:

{:ok, result} = instance.fetch_batch_for_polling("value")

Parameters:

Name	Type	Required	Description
`batch_id`	`String.t()`	Yes	The batch id

Returns: BatchObject

Errors: Returns {:error, reason}

wait_for_batch()

Poll a batch until it reaches a terminal status (Completed, Failed, Expired, Cancelled).

Uses exponential backoff with configurable initial interval, maximum interval, and backoff multiplier. Optionally supports a timeout that aborts polling if exceeded.

Errors:

Returns BatchWaitError.Failed if the batch reaches a failure terminal status. Returns BatchWaitError.Timeout if the configured timeout is exceeded. Returns BatchWaitError.Client for underlying client errors.

Signature:

def wait_for_batch(batch_id, config)

Example:

{:ok, result} = instance.wait_for_batch("value", %{{}})

Parameters:

Name	Type	Required	Description
`batch_id`	`String.t()`	Yes	The batch id
`config`	`WaitForBatchConfig`	Yes	The configuration options

Returns: BatchObject

Errors: Returns {:error, reason}

DeleteResponse

Response from a delete operation.

Field	Type	Default	Description
`id`	`String.t()`	—	ID of the deleted resource.
`object`	`String.t()`	—	Object type.
`deleted`	`boolean()`	—	Confirmation that the resource was deleted.

DeveloperMessage

Developer message (system-like message for Claude models).

Field	Type	Default	Description
`content`	`String.t()`	—	Developer-specific instructions or context.
`name`	`String.t() \| nil`	`nil`	Optional name for the developer message source.

DocumentContent

PDF/document content part for vision-capable models.

Field	Type	Default	Description
`data`	`String.t()`	—	Base64-encoded document data or URL.
`media_type`	`String.t()`	—	MIME type (e.g., “application/pdf”, “text/csv”).

EmbeddingObject

A single embedding vector.

Field	Type	Default	Description
`object`	`String.t()`	—	Always `"embedding"` from OpenAI-compatible APIs. Stored as a plain `String` so non-standard provider values do not break deserialization.
`embedding`	`list(float())`	—	The embedding vector. Providers may return this as a JSON float array or, when `encoding_format: "base64"` was requested, as a base64 string of little-endian `f32` bytes. Base64 responses are decoded on read; this field always serializes back out as a JSON float array.
`index`	`integer()`	—	Index in the batch (corresponds to input order).

EmbeddingRequest

Embedding request.

Field	Type	Default	Description
`model`	`String.t()`	—	Model ID (e.g., `"text-embedding-3-small"`).
`input`	`EmbeddingInput`	`:single`	Text or texts to embed.
`encoding_format`	`EmbeddingFormat \| nil`	`nil`	Output format: float (native) or base64.
`dimensions`	`integer() \| nil`	`nil`	Requested embedding dimensions (if supported by the model).
`user`	`String.t() \| nil`	`nil`	User identifier for request tracking.

EmbeddingResponse

Embedding response.

Field	Type	Default	Description
`object`	`String.t()`	—	Always `"list"` from OpenAI-compatible APIs. Stored as a plain `String` so non-standard provider values do not break deserialization.
`data`	`list(EmbeddingObject)`	—	List of embeddings.
`model`	`String.t()`	—	Model used to generate embeddings.
`usage`	`Usage \| nil`	`/* serde(default) */`	Token usage (input tokens only; embeddings have zero output tokens).

FileListQuery

Query parameters for listing files.

Field	Type	Default	Description
`purpose`	`String.t() \| nil`	`nil`	Filter by file purpose (e.g., `"batch"`, `"fine-tune"`).
`limit`	`integer() \| nil`	`nil`	Maximum number of results to return. Defaults to 20.
`after`	`String.t() \| nil`	`nil`	Pagination cursor: return results after this file ID.

FileListResponse

Response from listing files.

Field	Type	Default	Description
`object`	`String.t()`	—	Object type (always `"list"`).
`data`	`list(FileObject)`	`\[\]`	List of file objects.
`has_more`	`boolean() \| nil`	`nil`	Whether more results are available.

FileObject

An uploaded file object.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique file ID.
`object`	`String.t()`	—	Object type (always `"file"`).
`bytes`	`integer()`	—	File size in bytes.
`created_at`	`integer()`	—	Unix timestamp of file creation.
`filename`	`String.t()`	—	Filename.
`purpose`	`String.t()`	—	File purpose.
`status`	`String.t() \| nil`	`nil`	Processing status (e.g., `"uploaded"`, `"processed"`).

FunctionCall

Function call details.

Field	Type	Default	Description
`name`	`String.t()`	—	Function name.
`arguments`	`String.t()`	—	Arguments as a JSON string (parse with serde_json.from_str).

FunctionDefinition

Function definition exposed to the model.

Field	Type	Default	Description
`name`	`String.t()`	—	Name of the function. Required and must be alphanumeric + underscores.
`description`	`String.t() \| nil`	`/* serde(default) */`	Human-readable description explaining what the function does.
`parameters`	`term() \| nil`	`/* serde(default) */`	JSON Schema defining the function’s parameters.
`strict`	`boolean() \| nil`	`/* serde(default) */`	If true, enforce strict JSON schema validation for arguments.

FunctionMessage

Deprecated legacy function-role message body.

Field	Type	Default	Description
`content`	`String.t()`	—	The extracted text content
`name`	`String.t()`	—	The name

HealthChecker

Abstraction over a health probe strategy.

Implementors issue a lightweight probe against upstream (typically a provider base URL or named identifier) and report HealthStatus.

Functions

check()

Probe upstream and return its current HealthStatus.

The parameter is taken by value (String) so that implementations can move it into the returned future without a clone, making the 'static + Send bound on the future trivially satisfiable.

Signature:

def check(upstream)

Example:

{:ok, result} = instance.check("value")

Parameters:

Name	Type	Required	Description
`upstream`	`String.t()`	Yes	The upstream

Returns: HealthStatus

Image

A single generated image, returned as either a URL or base64 data.

Field	Type	Default	Description
`url`	`String.t() \| nil`	`nil`	Image URL (if response_format was “url”).
`b64_json`	`String.t() \| nil`	`nil`	Base64-encoded image data (if response_format was “b64_json”).
`revised_prompt`	`String.t() \| nil`	`nil`	The final prompt used to generate the image (DALL-E 3).

ImageUrl

An image URL reference with optional detail level for processing.

Field	Type	Default	Description
`url`	`String.t()`	—	URL of the image (data URI or HTTP/HTTPS URL).
`detail`	`ImageDetail \| nil`	`nil`	Detail level: low (512x512), high (2x2 tiles), or auto (model-selected).

ImagesResponse

Response containing generated images.

Field	Type	Default	Description
`created`	`integer()`	—	Unix timestamp of image creation.
`data`	`list(Image)`	`\[\]`	List of generated images.

IntentPrototype

An intent prototype: (intent_name, prototype_embedding, target_model_id).

Field	Type	Default	Description
`name`	`String.t()`	—	Human-readable name for the intent (used in logs/metrics).
`embedding`	`list(float())`	—	Pre-computed embedding vector for this intent.
`model`	`String.t()`	—	Model to route to when this intent is detected.

JsonSchemaFormat

JSON Schema specification for constrained output.

Field	Type	Default	Description
`name`	`String.t()`	—	Name of the schema (must be unique in the request).
`description`	`String.t() \| nil`	`nil`	Description of what the schema represents.
`schema`	`term()`	—	JSON Schema object defining the output structure.
`strict`	`boolean() \| nil`	`nil`	If true, enforce strict schema validation.

LlmBudgetConfig

Budget enforcement configuration.

Field	Type	Default	Description
`global_limit`	`float() \| nil`	`nil`	Global spend limit in USD.
`model_limits`	`map() \| nil`	`%{}`	Per-model spend limits in USD, keyed by model name.
`enforcement`	`String.t() \| nil`	`nil`	Enforcement mode: `"hard"` (reject over-budget requests) or `"soft"` (log only).

LlmCacheConfig

Response cache configuration.

Field	Type	Default	Description
`max_entries`	`integer() \| nil`	`nil`	Maximum number of cached entries.
`ttl_seconds`	`integer() \| nil`	`nil`	Cache entry time-to-live, in seconds.
`backend`	`String.t() \| nil`	`nil`	Cache backend name (e.g. `"memory"`, or an `opendal` scheme).
`backend_config`	`map() \| nil`	`%{}`	Backend-specific configuration key/value pairs.

LlmConfig

Canonical configuration for an LLM client.

All fields except model are optional so that partially-specified configs (e.g. from environment-driven defaults) round-trip cleanly. Convert to a runtime client configuration via LlmConfig.into_client_builder.

temperature and max_tokens are request-time parameters rather than client-level settings; they are carried on this struct for callers to read when building individual requests, and are intentionally not mapped by LlmConfig.into_client_builder.

Field	Type	Default	Description
`model`	`String.t()`	—	Model identifier (e.g. `"gpt-4o"`, `"bedrock/anthropic.claude-3-sonnet-20240229-v1:0"`).
`api_key`	`String.t() \| nil`	`nil`	API key for authentication.
`base_url`	`String.t() \| nil`	`nil`	Override base URL. When set, all requests go here and provider auto-detection is skipped.
`timeout_secs`	`integer() \| nil`	`nil`	Request timeout, in seconds.
`max_retries`	`integer() \| nil`	`nil`	Maximum number of retries on 429 / 5xx responses.
`temperature`	`float() \| nil`	`nil`	Sampling temperature for requests built from this config.
`max_tokens`	`integer() \| nil`	`nil`	Maximum number of tokens to generate for requests built from this config.
`load_env`	`boolean() \| nil`	`nil`	Automatically load the API key from the provider’s environment variable when no explicit key is provided (default: `true`).
`headers`	`map() \| nil`	`%{}`	Extra headers sent on every request.
`providers`	`list(LlmProviderConfig) \| nil`	`\[\]`	Custom provider configurations, in addition to the built-in providers.
`cache`	`LlmCacheConfig \| nil`	`nil`	Response cache configuration.
`budget`	`LlmBudgetConfig \| nil`	`nil`	Budget enforcement configuration.
`rate_limit`	`LlmRateLimitConfig \| nil`	`nil`	Per-model rate limiting configuration.
`cost_tracking`	`boolean() \| nil`	`nil`	Enable per-request cost tracking.
`tracing`	`boolean() \| nil`	`nil`	Enable OpenTelemetry-compatible tracing spans.
`cooldown_secs`	`integer() \| nil`	`nil`	Cooldown duration after transient errors, in seconds.
`health_check_secs`	`integer() \| nil`	`nil`	Background health check interval, in seconds.
`bedrock`	`BedrockConfig \| nil`	`nil`	AWS Bedrock configuration (region, credentials, cross-region routing).

Functions

providers()

Get the custom provider configurations from this config.

Signature:

def providers()

Example:

{:ok, result} = instance.providers()

Returns: list(LlmProviderConfig)

LlmProviderConfig

A custom provider configuration entry.

Field	Type	Default	Description
`name`	`String.t()`	—	Provider name, used to key model prefix matching.
`base_url`	`String.t()`	—	Base URL for the provider’s OpenAI-compatible API.
`auth_header`	`String.t() \| nil`	`nil`	Header name used to carry the API key (defaults to `Authorization` when unset).
`model_prefixes`	`list(String.t())`	`\[\]`	Model name prefixes routed to this provider (e.g. `\["my-provider/"\]`).

LlmRateLimitConfig

Per-model rate limiting configuration.

Field	Type	Default	Description
`rpm`	`integer() \| nil`	`nil`	Requests per minute limit.
`tpm`	`integer() \| nil`	`nil`	Tokens per minute limit.
`window_seconds`	`integer() \| nil`	`nil`	Rate limit window, in seconds.

ModelInfo

Public, FFI-friendly snapshot of a model’s pricing and capability metadata, projected from ModelPricing.

Unlike ModelPricing (which is excluded from binding generation), ModelInfo is an owned plain-data DTO safe to hand across the FFI boundary — see model_info.

Field	Type	Default	Description
`input_cost_per_token`	`float()`	—	Cost in USD per input (prompt) token.
`output_cost_per_token`	`float()`	—	Cost in USD per output (completion) token.
`cache_read_input_token_cost`	`float() \| nil`	`nil`	Cost in USD per cached input token (cache hit / read).
`cache_creation_input_token_cost`	`float() \| nil`	`nil`	Cost in USD per token written to the prompt cache.
`input_cost_per_audio_token`	`float() \| nil`	`nil`	Cost in USD per input audio token.
`output_cost_per_audio_token`	`float() \| nil`	`nil`	Cost in USD per output audio token.
`output_cost_per_reasoning_token`	`float() \| nil`	`nil`	Cost in USD per reasoning (extended-thinking) output token.
`max_tokens`	`integer() \| nil`	`nil`	Total context window size in tokens (input + output).
`max_input_tokens`	`integer() \| nil`	`nil`	Maximum input (prompt) tokens accepted.
`max_output_tokens`	`integer() \| nil`	`nil`	Maximum output (completion) tokens the model can generate.
`mode`	`String.t() \| nil`	`nil`	Best-effort operating mode, e.g. `"chat"`, `"embedding"`.
`supports_vision`	`boolean() \| nil`	`nil`	The model accepts image input.
`supports_function_calling`	`boolean() \| nil`	`nil`	The model supports tool / function calling.
`supports_reasoning`	`boolean() \| nil`	`nil`	The model supports extended-thinking / reasoning tokens.
`supports_structured_output`	`boolean() \| nil`	`nil`	The model supports JSON-mode or `response_format` structured output.
`supports_audio_input`	`boolean() \| nil`	`nil`	The model accepts audio input.
`supports_audio_output`	`boolean() \| nil`	`nil`	The model can generate audio output.
`supports_prompt_caching`	`boolean() \| nil`	`nil`	The model supports prompt caching.
`tiers`	`list(ModelTier)`	`\[\]`	Context-tiered pricing overrides, sorted by ascending `min_context_tokens`. Empty when the model has flat pricing.

ModelObject

A model available from the API.

Field	Type	Default	Description
`id`	`String.t()`	—	Model ID (e.g., `"gpt-4o"`, `"claude-3-5-sonnet"`).
`object`	`String.t()`	—	Always `"model"` from OpenAI-compatible APIs. Stored as a plain `String` so non-standard provider values do not break deserialization. Defaults to empty when a provider omits the field.
`created`	`integer()`	—	Unix timestamp of model creation (or release date). Defaults to `0` when a provider omits it — DeepSeek and some other OpenAI-compatible providers do not return `created` from `/v1/models`.
`owned_by`	`String.t()`	—	Organization or entity that owns the model. Defaults to empty when a provider omits the field.

ModelTier

Public, FFI-friendly snapshot of a single context-window pricing tier, projected from PricingTier.

Field	Type	Default	Description
`min_context_tokens`	`integer()`	—	The tier applies when the prompt/context token count is at least this value.
`input_cost_per_token`	`float()`	—	Cost in USD per input (prompt) token within this tier.
`output_cost_per_token`	`float()`	—	Cost in USD per output (completion) token within this tier.
`cache_read_input_token_cost`	`float() \| nil`	`nil`	Cost in USD per cached input token within this tier.
`cache_creation_input_token_cost`	`float() \| nil`	`nil`	Cost in USD per cache-write token within this tier.
`input_cost_per_audio_token`	`float() \| nil`	`nil`	Cost in USD per input audio token within this tier.
`output_cost_per_audio_token`	`float() \| nil`	`nil`	Cost in USD per output audio token within this tier.
`output_cost_per_reasoning_token`	`float() \| nil`	`nil`	Cost in USD per reasoning output token within this tier.

ModelsListResponse

Response listing available models.

Field	Type	Default	Description
`object`	`String.t()`	—	Always `"list"` from OpenAI-compatible APIs. Stored as a plain `String` so non-standard provider values do not break deserialization. Defaults to empty when a provider omits the field.
`data`	`list(ModelObject)`	`\[\]`	List of available models.

ModerationCategories

Boolean flags for each moderation category.

Field	Type	Default	Description
`sexual`	`boolean()`	—	Sexual content.
`hate`	`boolean()`	—	Hate speech.
`harassment`	`boolean()`	—	Harassment.
`self_harm`	`boolean()`	—	Self-harm content.
`sexual_minors`	`boolean()`	—	Sexual content involving minors.
`hate_threatening`	`boolean()`	—	Hate speech that threatens violence.
`violence_graphic`	`boolean()`	—	Graphic violence.
`self_harm_intent`	`boolean()`	—	Intent to self-harm.
`self_harm_instructions`	`boolean()`	—	Instructions for self-harm.
`harassment_threatening`	`boolean()`	—	Harassment that threatens violence.
`violence`	`boolean()`	—	Non-graphic violence.

ModerationCategoryScores

Confidence scores for each moderation category.

Field	Type	Default	Description
`sexual`	`float()`	—	Sexual content score.
`hate`	`float()`	—	Hate speech score.
`harassment`	`float()`	—	Harassment score.
`self_harm`	`float()`	—	Self-harm content score.
`sexual_minors`	`float()`	—	Sexual content involving minors score.
`hate_threatening`	`float()`	—	Hate speech that threatens violence score.
`violence_graphic`	`float()`	—	Graphic violence score.
`self_harm_intent`	`float()`	—	Intent to self-harm score.
`self_harm_instructions`	`float()`	—	Instructions for self-harm score.
`harassment_threatening`	`float()`	—	Harassment that threatens violence score.
`violence`	`float()`	—	Non-graphic violence score.

ModerationRequest

Request to classify content for policy violations.

Field	Type	Default	Description
`input`	`ModerationInput`	`:single`	Text or texts to check.
`model`	`String.t() \| nil`	`nil`	Model ID (e.g., `"text-moderation-latest"`). Optional; API uses default if unset.

ModerationResponse

Response from the moderation endpoint.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique identifier for this moderation request.
`model`	`String.t()`	—	Model used for classification.
`results`	`list(ModerationResult)`	—	Results for each input string.

ModerationResult

A single moderation classification result.

Field	Type	Default	Description
`flagged`	`boolean()`	—	True if any category was flagged.
`categories`	`ModerationCategories`	—	Boolean flags for each moderation category.
`category_scores`	`ModerationCategoryScores`	—	Confidence scores for each category.

OcrImage

An image extracted from an OCR page.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique image identifier within the document.
`image_base64`	`String.t() \| nil`	`/* serde(default) */`	Base64-encoded image data (if `include_image_base64` was true).

OcrPage

A single page of OCR output.

Field	Type	Default	Description
`index`	`integer()`	—	Page index (0-based).
`markdown`	`String.t()`	—	Extracted page content as Markdown.
`images`	`list(OcrImage) \| nil`	`/* serde(default) */`	Embedded images extracted from the page (if `include_image_base64` was true).
`dimensions`	`PageDimensions \| nil`	`/* serde(default) */`	Page dimensions in pixels, if available.

OcrRequest

An OCR request.

Field	Type	Default	Description
`model`	`String.t()`	—	The model/provider to use (e.g. `"mistral/mistral-ocr-latest"`).
`document`	`OcrDocument`	`:url`	The document to process (URL or base64).
`pages`	`list(integer()) \| nil`	`\[\]`	Specific pages to process (1-indexed). `nil` means all pages.
`include_image_base64`	`boolean() \| nil`	`nil`	Whether to include base64-encoded images of each processed page.

OcrResponse

An OCR response.

Field	Type	Default	Description
`pages`	`list(OcrPage)`	—	Extracted pages in order.
`model`	`String.t()`	—	Model/provider used for OCR.
`usage`	`Usage \| nil`	`/* serde(default) */`	Token usage, if reported by the provider.

PageDimensions

Page dimensions in pixels.

Field	Type	Default	Description
`width`	`integer()`	—	Width in pixels.
`height`	`integer()`	—	Height in pixels.

PromptTokensDetails

Breakdown of tokens used in the prompt portion of a request.

cached_tokens is included in Usage.prompt_tokens — it is not an additional charge on top of the prompt token count. When pricing supports a cache_read_input_token_cost, the cached portion is billed at the discounted rate and the remainder at the regular input rate.

Field	Type	Default	Description
`cached_tokens`	`integer()`	—	Cached tokens present in the prompt. Defaults to 0 when absent.
`audio_tokens`	`integer()`	—	Audio input tokens present in the prompt. Defaults to 0 when absent.

ProviderCapabilities

Static capability flags for a provider.

Each flag indicates whether the provider’s models generally support that feature. For providers that aggregate many underlying models (e.g. Bedrock, OpenRouter, vLLM) the flags reflect the superset of available model capabilities — a flag being true means at least one model supports the feature, not every model.

All flags default to false so that newly added providers are safe.

Access via the crate-level capabilities function:

Field	Type	Default	Description
`vision`	`boolean()`	—	The provider accepts image input in chat messages.
`reasoning`	`boolean()`	—	The provider supports extended-thinking / reasoning tokens.
`structured_output`	`boolean()`	—	The provider supports JSON-mode or `response_format` structured output.
`function_calling`	`boolean()`	—	The provider supports tool / function calling.
`audio_in`	`boolean()`	—	The provider accepts audio as input.
`audio_out`	`boolean()`	—	The provider can generate audio / TTS output.
`video_in`	`boolean()`	—	The provider accepts video as input.

ProviderConfig

Static configuration for a single provider entry in providers.json.

This struct deliberately does not include capability flags or streaming format, which are accessed via the capabilities function.

Field	Type	Default	Description
`name`	`String.t()`	—	Provider identifier (matches the entry key in providers.json).
`display_name`	`String.t() \| nil`	`nil`	Human-readable provider name shown in UIs.
`base_url`	`String.t() \| nil`	`nil`	Base URL used as the default for this provider’s HTTP client.
`auth`	`AuthConfig \| nil`	`nil`	Authentication scheme metadata (auth type + env var holding the key).
`endpoints`	`list(String.t()) \| nil`	`nil`	Supported endpoint kinds (e.g. `chat`, `embeddings`).
`model_prefixes`	`list(String.t()) \| nil`	`nil`	Model-name prefixes claimed by this provider (e.g. `\["gpt-", "o1-"\]`).
`param_mappings`	`map() \| nil`	`nil`	Parameter key renaming for this provider. Each entry maps an OpenAI-spec field name (e.g. `"max_completion_tokens"`) to the name this provider expects (e.g. `"max_tokens"`). Applied automatically by `ConfigDrivenProvider.transform_request`.

RateLimitConfig

Configuration for per-model rate limits.

Field	Type	Default	Description
`rpm`	`integer() \| nil`	`nil`	Maximum requests per window. `nil` means unlimited.
`tpm`	`integer() \| nil`	`nil`	Maximum tokens per window. `nil` means unlimited.
`window`	`integer()`	`60000ms`	Fixed window duration (defaults to 60 s).

Functions

default()

Signature:

def default()

Example:

{:ok, result} = RateLimitConfig.default()

Returns: RateLimitConfig

RerankRequest

Request to rerank documents by relevance to a query.

Field	Type	Default	Description
`model`	`String.t()`	—	Model ID (e.g., `"cohere/rerank-english-v3.0"`).
`query`	`String.t()`	—	The search query.
`documents`	`list(RerankDocument)`	`\[\]`	Documents to rerank.
`top_n`	`integer() \| nil`	`nil`	Return only the top N results. Optional.
`return_documents`	`boolean() \| nil`	`nil`	Include the document content in results. Defaults to false.

RerankResponse

Response from the rerank endpoint.

Field	Type	Default	Description
`id`	`String.t() \| nil`	`nil`	Unique identifier for this rerank request.
`results`	`list(RerankResult)`	—	Reranked documents in order of relevance.
`meta`	`term() \| nil`	`/* serde(default) */`	Optional metadata about the reranking operation.

RerankResult

A single reranked document with its relevance score.

Field	Type	Default	Description
`index`	`integer()`	—	Original document index in the input list.
`relevance_score`	`float()`	—	Relevance score in `\[0, 1\]`. Higher indicates more relevant.
`document`	`RerankResultDocument \| nil`	`/* serde(default) */`	Original document content (if `return_documents` was true).

RerankResultDocument

The text content of a reranked document, returned when return_documents is true.

Field	Type	Default	Description
`text`	`String.t()`	—	Document text.

ResponseObject

Response from a structured response request.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique response ID.
`object`	`String.t()`	—	Object type (e.g., `"response"`).
`created_at`	`integer()`	—	Unix timestamp of response creation.
`model`	`String.t()`	—	Model used to generate the response.
`status`	`String.t()`	—	Status (e.g., `"succeeded"`, `"failed"`).
`output`	`list(ResponseOutputItem)`	`\[\]`	Output items from the response.
`usage`	`ResponseUsage \| nil`	`nil`	Token usage.
`error`	`term() \| nil`	`nil`	Error details (if status is “failed”).

ResponseOutputItem

A single output item from the response.

Field	Type	Default	Description
`item_type`	`String.t()`	—	Output type (e.g., `"text"`, `"object"`, `"error"`).
`content`	`term()`	—	Output content (flattened into the object).

ResponseTool

A tool available for the response request.

Field	Type	Default	Description
`tool_type`	`String.t()`	—	Tool type (e.g., “extractor”, “search”).
`config`	`term()`	—	Tool configuration (flattened into the object).

ResponseUsage

Token usage for a response.

Field	Type	Default	Description
`input_tokens`	`integer()`	—	Input tokens used.
`output_tokens`	`integer()`	—	Output tokens used.
`total_tokens`	`integer()`	—	Total tokens used.

SearchRequest

A search request.

Field	Type	Default	Description
`model`	`String.t()`	—	The model/provider to use (e.g. `"brave/web-search"`, `"tavily/search"`).
`query`	`String.t()`	—	The search query string.
`max_results`	`integer() \| nil`	`nil`	Maximum number of results to return.
`search_domain_filter`	`list(String.t()) \| nil`	`\[\]`	Domain filter — restrict results to specific domains.
`country`	`String.t() \| nil`	`nil`	Country code for localized results (ISO 3166-1 alpha-2, e.g., `"US"`, `"FR"`).

SearchResponse

A search response.

Field	Type	Default	Description
`results`	`list(SearchResult)`	—	List of search results.
`model`	`String.t()`	—	Model/provider that performed the search.

SearchResult

An individual search result.

Field	Type	Default	Description
`title`	`String.t()`	—	Result title.
`url`	`String.t()`	—	Result URL.
`snippet`	`String.t()`	—	Text snippet or excerpt from the page.
`date`	`String.t() \| nil`	`/* serde(default) */`	Publication or last-updated date, if available.

SingleflightResult

The value broadcast from a singleflight leader to all followers.

The error value is shared so every follower receives the same upstream failure without cloning the underlying error.

SpecificFunction

Name of the specific function to invoke.

Field	Type	Default	Description
`name`	`String.t()`	—	Function name.

SpecificToolChoice

Directive to call a specific tool.

Field	Type	Default	Description
`choice_type`	`ToolType`	`:function`	Tool type (always “function”).
`function`	`SpecificFunction`	—	The specific function to invoke.

StreamChoice

A streaming choice with incremental delta.

Field	Type	Default	Description
`index`	`integer()`	—	Index of this choice in the choices array.
`delta`	`StreamDelta`	—	Incremental update to the message (content, tool calls, etc.).
`finish_reason`	`FinishReason \| nil`	`nil`	Why the stream ended (present only in final chunk).

StreamDelta

Incremental delta in a stream chunk.

Field	Type	Default	Description
`role`	`String.t() \| nil`	`nil`	Role (typically present only in the first chunk).
`content`	`String.t() \| nil`	`nil`	Partial content chunk (e.g., a few words of the response).
`tool_calls`	`list(StreamToolCall) \| nil`	`\[\]`	Partial tool calls being streamed.
`function_call`	`StreamFunctionCall \| nil`	`nil`	Deprecated legacy function_call delta; retained for API compatibility.
`refusal`	`String.t() \| nil`	`nil`	Partial refusal message.
`reasoning_content`	`String.t() \| nil`	`nil`	Partial reasoning/thinking tokens (OpenAI-compatible extension used by DeepSeek R1, Qwen, etc.).

StreamFunctionCall

Partial function call details in a stream.

Field	Type	Default	Description
`name`	`String.t() \| nil`	`nil`	Function name (typically in the first chunk).
`arguments`	`String.t() \| nil`	`nil`	Partial JSON arguments chunk.

StreamOptions

Options for streaming responses.

Field	Type	Default	Description
`include_usage`	`boolean() \| nil`	`nil`	If true, include token usage in the final stream chunk.

StreamToolCall

A streaming tool call being built incrementally.

Field	Type	Default	Description
`index`	`integer()`	—	Index of this tool call in the tool_calls array.
`id`	`String.t() \| nil`	`nil`	Tool call ID (typically in the first chunk for this call).
`call_type`	`ToolType \| nil`	`nil`	Tool type (typically “function”).
`function`	`StreamFunctionCall \| nil`	`nil`	Partial function name and arguments.

SystemMessage

System message guiding model behavior for the entire conversation.

Field	Type	Default	Description
`content`	`UserContent`	`:text`	Instructions or context that apply throughout the conversation. Accepts either a plain text string or an array of content parts, mirroring `UserContent` so that `Message.system_with_parts` works.
`name`	`String.t() \| nil`	`nil`	Optional name for the system message source.

ToolCall

A tool call the model wants to execute.

Field	Type	Default	Description
`id`	`String.t()`	—	Unique ID for this call, used to reference in tool result messages.
`call_type`	`ToolType`	—	Tool type (always “function”).
`function`	`FunctionCall`	—	Function name and arguments.

ToolMessage

Tool execution result returned to the model.

Field	Type	Default	Description
`content`	`UserContent`	`:text`	Result of the tool execution as plain text or an array of content parts (text, images, documents, audio), mirroring `UserMessage.content`. `#\[serde(untagged)\]` on `UserContent` means a bare JSON string still deserialises into `Text`, so tool results persisted before this field carried structured content continue to round-trip.
`tool_call_id`	`String.t()`	—	ID of the tool call this result responds to.
`name`	`String.t() \| nil`	`nil`	Optional tool/function name.

TranscriptionResponse

Response from a transcription request.

Field	Type	Default	Description
`text`	`String.t()`	—	The transcribed text.
`language`	`String.t() \| nil`	`nil`	Detected language (ISO-639-1 code).
`duration`	`float() \| nil`	`nil`	Total audio duration in seconds.
`segments`	`list(TranscriptionSegment) \| nil`	`\[\]`	Detailed segment-level transcription (if response_format is “verbose_json”).

TranscriptionSegment

A segment of transcribed audio with timing information.

Field	Type	Default	Description
`id`	`integer()`	—	Segment index (0-based).
`start`	`float()`	—	Start time in seconds.
`end`	`float()`	—	End time in seconds.
`text`	`String.t()`	—	Transcribed text for this segment.

Usage

Token-usage accounting returned by the provider on each completion / embedding call.

Field	Type	Default	Description
`prompt_tokens`	`integer()`	—	Prompt tokens used. Defaults to 0 when absent (some providers omit this).
`completion_tokens`	`integer()`	—	Completion tokens used. Defaults to 0 when absent (e.g. embedding responses).
`total_tokens`	`integer()`	—	Total tokens used. Defaults to 0 when absent (some providers omit this).
`prompt_tokens_details`	`PromptTokensDetails \| nil`	`nil`	Breakdown of tokens used in the prompt, including cached tokens served at the provider’s discounted cache-read rate. Absent when the provider does not return prompt-token details.

UserMessage

User message in the conversation.

Field	Type	Default	Description
`content`	`UserContent`	`:text`	Message content as plain text or array of content parts (text, images, documents, audio).
`name`	`String.t() \| nil`	`nil`	Optional name for the user.

WaitForBatchConfig

Configuration for polling a batch until terminal status.

All time values are in seconds as f64 so the struct bridges across FFI boundaries without requiring a Duration shim.

Field	Type	Default	Description
`initial_interval_secs`	`float()`	`5`	Initial interval between polls, in seconds.
`max_interval_secs`	`float()`	`60`	Maximum interval between polls (backoff plateau), in seconds.
`backoff_multiplier`	`float()`	`1.5`	Exponential backoff multiplier (e.g., 1.5 increases delay by 50% each poll).
`timeout_secs`	`float() \| nil`	`nil`	Optional timeout in seconds — polling fails if this duration is exceeded.

Functions

default()

Signature:

def default()

Example:

{:ok, result} = WaitForBatchConfig.default()

Returns: WaitForBatchConfig

Enums

Message

A chat message in a conversation.

Value	Description
`system`	System — Fields: `0`: `SystemMessage`
`user`	User — Fields: `0`: `UserMessage`
`assistant`	Assistant — Fields: `0`: `AssistantMessage`
`tool`	Tool — Fields: `0`: `ToolMessage`
`developer`	Developer — Fields: `0`: `DeveloperMessage`
`function`	Deprecated legacy function-role message; retained for API compatibility. — Fields: `0`: `FunctionMessage`

UserContent

User message content as either plain text or a list of multimodal parts.

Value	Description
`text`	Plain text content. — Fields: `0`: `String.t()`
`parts`	Array of content parts (text, images, documents, audio). — Fields: `0`: `list(ContentPart)`

ContentPart

A single content part in a user message — text, image, document, or audio.

Value	Description
`text`	Plain text. — Fields: `text`: `String.t()`
`image_url`	Image identified by URL (with optional detail level). — Fields: `image_url`: `ImageUrl`
`document`	Document file (PDF, CSV, etc.) as base64 or URL. — Fields: `document`: `DocumentContent`
`input_audio`	Audio input as base64. — Fields: `input_audio`: `AudioContent`

ImageDetail

Image detail level controlling token cost and processing.

Value	Description
`low`	Low detail: scales image to 512x512, uses fewer tokens.
`high`	High detail: processes up to 2x2 grid of tiles, higher token cost.
`auto`	Auto: model chooses low or high based on image dimensions.

AssistantContent

Content shape for assistant messages.

#[serde(untagged)] means providers returning a plain scalar string for the content field still deserialise correctly into AssistantContent.Text(_). Providers returning an array of typed parts (e.g. after an image-generation or audio-synthesis request) deserialise into AssistantContent.Parts(_).

Value	Description
`text`	Plain text response (the common case for text-only models). — Fields: `0`: `String.t()`
`parts`	Structured parts — text, refusals, output images, output audio. — Fields: `0`: `list(AssistantPart)`

AssistantPart

One part of a structured assistant response.

#[serde(tag = "type", rename_all = "snake_case")] matches OpenAI’s parts-spec discriminator ("type": "text", "type": "output_image", …).

Value	Description
`text`	A text segment of the response. — Fields: `text`: `String.t()`
`refusal`	A refusal — the model declined to respond. — Fields: `refusal`: `String.t()`
`output_image`	An image produced by the model (e.g. `gpt-image-1`, Gemini Imagen). — Fields: `image_url`: `ImageUrl`
`output_audio`	Audio produced by the model (e.g. `gpt-4o-audio-preview`). — Fields: `audio`: `AudioContent`

ToolType

The type discriminator for tool/tool-call objects.

Per the OpenAI spec this is always "function". Using an enum enforces that constraint at the type level and rejects any other value on deserialization.

Value	Description
`function`	Function

ToolChoice

Tool usage mode or a specific tool to call.

Value	Description
`mode`	Predefined mode: auto, required, or none. — Fields: `0`: `ToolChoiceMode`
`specific`	Force a specific tool to be called. — Fields: `0`: `SpecificToolChoice`

ToolChoiceMode

Tool choice mode.

Value	Description
`auto`	Model may or may not call tools; default behavior.
`required`	Model must call at least one tool.
`none`	Model must not call any tools.

ResponseFormat

Wire format for the chat completions response_format field.

Provider mapping

OpenAI (and OpenAI-compatible providers): emitted verbatim as {"type": "json_schema", "json_schema": {...}} per the chat-completions spec.
Gemini / Vertex AI: translated to generationConfig.responseMimeType = "application/json" and generationConfig.responseSchema = <schema>. The name, description, and strict fields are dropped — Gemini’s structured-output API does not consume them.
Anthropic: no native JSON mode. A system instruction is prepended asking the model to respond with valid JSON. strict is advisory only; callers should still validate the returned JSON if the schema is load-bearing.

Value	Description
`text`	Plain text output (default).
`json_object`	Output must be valid JSON object (no schema validation).
`json_schema`	Output must conform to the specified JSON schema. — Fields: `json_schema`: `JsonSchemaFormat`

StopSequence

Stop sequence(s) that cause the model to stop generating.

Value	Description
`single`	Single stop sequence. — Fields: `0`: `String.t()`
`multiple`	Multiple stop sequences. — Fields: `0`: `list(String.t())`

Modality

Output modality requested from the model.

Passed as modalities: ["text", "audio"] (OpenAI) or translated to generationConfig.responseModalities (Gemini / Vertex AI).

Value	Description
`text`	Text output (the default for all providers).
`audio`	Audio / speech output.
`image`	Image output (Gemini Imagen, gpt-image-1).

FinishReason

Why a choice stopped generating tokens.

Value	Description
`stop`	Stop
`length`	Length
`tool_calls`	Tool calls
`content_filter`	Content filter
`function_call`	Deprecated legacy finish reason; retained for API compatibility.
`other`	Catch-all for unknown finish reasons returned by non-OpenAI providers. Note: this intentionally does not carry the original string (e.g. `Other(String)`). Using `#\[serde(other)\]` requires a unit variant, and switching to `#\[serde(untagged)\]` would change deserialization semantics for all variants. The original value can be recovered by inspecting the raw JSON if needed.

ReasoningEffort

Controls how much reasoning effort the model should use.

Value	Description
`low`	Low
`medium`	Medium
`high`	High
`minimal`	Minimal
`max`	Max

EmbeddingFormat

The format in which the embedding vectors are returned.

Value	Description
`float`	32-bit floating-point numbers (default).
`base64`	Base64-encoded string representation of the floats.

EmbeddingInput

Text or texts to embed.

Value	Description
`single`	Single text string. — Fields: `0`: `String.t()`
`multiple`	Multiple text strings (batch embedding). — Fields: `0`: `list(String.t())`

ModerationInput

Input to the moderation endpoint — a single string or multiple strings.

Value	Description
`single`	Single text string. — Fields: `0`: `String.t()`
`multiple`	Multiple text strings (batch moderation). — Fields: `0`: `list(String.t())`

RerankDocument

A document to be reranked — either a plain string or an object with a text field.

Value	Description
`text`	Plain text document content. — Fields: `0`: `String.t()`
`object`	Document with explicit text field (may include metadata). — Fields: `text`: `String.t()`

OcrDocument

Document input for OCR — either a URL or inline base64 data.

Value	Description
`url`	A publicly accessible document URL. — Fields: `url`: `String.t()`
`base64`	Inline base64-encoded document data. — Fields: `data`: `String.t()`, `media_type`: `String.t()`

FilePurpose

Purpose of an uploaded file.

Value	Description
`assistants`	File for use with Assistants API.
`batch`	File for batch processing.
`fine_tune`	File for fine-tuning.
`vision`	File for vision/image tasks.

BatchStatus

Status of a batch job.

Value	Description
`validating`	Validating the input file.
`failed`	Job failed.
`in_progress`	Job is running.
`finalizing`	Finalizing results.
`completed`	Job completed successfully.
`expired`	Job expired before completion.
`cancelling`	Job is being cancelled.
`cancelled`	Job has been cancelled.

AuthHeaderFormat

How the API key is sent in the HTTP request.

Value	Description
`bearer`	Bearer token: `Authorization: Bearer <key>`
`api_key`	Custom header: e.g., `X-Api-Key: <key>` — Fields: `0`: `String.t()`
`none`	No authentication required.

StreamFormat

The streaming wire format a provider uses for its response stream.

Most providers use standard Server-Sent Events (SSE). AWS Bedrock uses a proprietary binary EventStream framing.

Deserialized from the streaming_format JSON field via serde.

Value	Description
`sse`	Standard Server-Sent Events (text/event-stream).
`aws_event_stream`	AWS EventStream binary framing (application/vnd.amazon.eventstream).

AuthType

Auth scheme used by a provider.

Value	Description
`bearer`	Standard `Authorization: Bearer <key>` header.
`api_key`	`x-api-key: <key>` header (also handles `"header"` and `"x-api-key"` aliases).
`none`	No authentication header required.
`unknown`	Unrecognised auth scheme — falls back to bearer.

Enforcement

How budget limits are enforced.

Value	Description
`hard`	Reject requests that would exceed the budget with `LiterLlmError.BudgetExceeded`.
`soft`	Allow requests through but emit a `tracing.warn!` when the budget is exceeded.

CacheBackend

Storage backend for the response cache.

Value	Description
`memory`	In-memory LRU cache (default). No external dependencies.
`open_dal`	OpenDAL-backed storage. Supports 40+ backends (S3, Redis, GCS, local FS, etc.). — Fields: `scheme`: `String.t()`, `config`: `map()`

CircuitState

Observable state of a circuit breaker.

Value	Description
`closed`	Requests flow through normally.
`open`	All requests are rejected; the circuit is waiting for the backoff to elapse.
`half_open`	One probe request is allowed through to test service health.

HealthStatus

The result of a single health probe.

Value	Description
`healthy`	The probe succeeded; the upstream is reachable.
`unhealthy`	The probe failed; the upstream may be down.

RefreshOutcome

Result of a refresh_catalog call.

Value	Description
`disabled`	`config.enabled` was `false`; no network, filesystem, or overlay activity occurred.
`from_cache`	The on-disk cache was fresh (age < `ttl_seconds`); the overlay was installed from the cached file without a network request.
`fetched`	The catalog was fetched over the network, the cache file was (best-effort) refreshed, and the overlay was installed from the fetched catalog.

Errors

LiterLlmError

All errors that can occur when using liter-llm.

Variant	Description
`authentication`	`status` preserves the exact HTTP status code received (401 or 403).
`rate_limited`	rate limited: {message}
`bad_request`	`status` preserves the exact HTTP status code received (400, 405, 413, 422, …).
`context_window_exceeded`	context window exceeded: {message}
`content_policy`	content policy violation: {message}
`not_found`	not found: {message}
`server_error`	`status` preserves the exact HTTP status code received (500, or other 5xx not covered by `ServiceUnavailable`).
`service_unavailable`	`status` preserves the exact HTTP status code received (502, 503, or 504).
`timeout`	request timeout
`streaming`	A catch-all for errors that occur during streaming response processing. This variant covers multiple sub-conditions including UTF-8 decoding failures, CRC/checksum mismatches (AWS EventStream), JSON parse errors in individual SSE chunks, and buffer overflow conditions. The `message` field contains a human-readable description of the specific failure.
`endpoint_not_supported`	provider {provider} does not support {endpoint}
`invalid_header`	invalid header {name:?}: {reason}
`serialization`	serialization error: {0}
`budget_exceeded`	budget exceeded: {message}
`hook_rejected`	hook rejected: {message}
`internal_error`	An internal logic error (e.g. unexpected Tower response variant). This should never surface in normal operation — if it does, it indicates a bug in the library.
`outbound_forbidden`	An outbound request was blocked by the active `OutboundPolicy`. Returned when `register_custom_provider` is called with a `base_url` that violates the policy (e.g. a private-range IP under `DenyPrivate`), or when the per-connection DNS resolver detects a forbidden address at connect time.
`idempotency_conflict`	A different request body was submitted for an existing `Idempotency-Key`. Per the OpenAI `Idempotency-Key` convention, once a key is used with a particular request body, subsequent requests using the same key must carry an identical body. A body mismatch is a hard error (not retryable). HTTP equivalent: 409 Conflict.
`idempotency_in_flight`	The same `Idempotency-Key` is already in-flight (another request with the same key is currently being processed). The caller should wait briefly and retry. The response is not yet available, and this request has been short-circuited to avoid running the operation twice. HTTP equivalent: 409 Conflict (retryable after a brief delay).

CatalogRefreshError

Errors from refresh_catalog and install_catalog_overlay_from_str.

On every variant, the overlay registry is left untouched: a previously installed overlay (or the embedded catalog, if none was ever installed) remains active. This is the air-gap-safety contract — a failed refresh never degrades pricing/model-info availability.

Variant	Description
`disabled`	Runtime catalog refresh was not enabled. `refresh_catalog` itself never returns this — it returns `Ok(RefreshOutcome.Disabled)` instead — but the variant is part of the public error surface for callers that want to treat “disabled” as a hard error.
`insecure_url`	`source_url` did not use the `https` scheme, or failed to parse as a URL at all. There is no host allowlist: the URL is user-configurable for self-hosted catalog mirrors, so only the scheme is enforced.
`fetch`	The network fetch failed, timed out, or returned a non-success status.
`parse`	The fetched or cached catalog JSON failed to parse.
`cache`	A cache file read failed on an otherwise-fresh cache file. Cache write failures are best-effort and never surface as this error (see `refresh_catalog`).