Crate tt_shared 

Shared types, errors, and traits for the TokenTrimmer workspace.

No business logic lives here — only the contracts that other crates implement. See docs/02-provider-adapter-guide.md for the Provider trait.

Re-exports

pub use capability_check::message_text_for_estimation;

pub use capability_check::RequiredCapabilities;

pub use context::CallerTier;

pub use context::RequestContext;

pub use error::ProviderError;

pub use messages::parse_cache_control;

pub use messages::CacheControlConfig;

pub use messages::CacheMode;

pub use messages::ChatCompletionChunk;

pub use messages::ChatCompletionRequest;

pub use messages::ChatCompletionResponse;

pub use messages::Choice;

pub use messages::ContentPart;

pub use messages::EmbeddingsRequest;

pub use messages::EmbeddingsResponse;

pub use messages::Message;

pub use messages::MessageContent;

pub use messages::Tool;

pub use messages::ToolCall;

pub use messages::ToolChoice;

pub use pricing::ModelInfo;

pub use pricing::ModelPricing;

pub use provider::Provider;

pub use url_guard::filter_extra_headers;

pub use url_guard::find_denied_header;

pub use url_guard::validate_provider_url;

pub use url_guard::UrlGuardError;

pub use usage::Usage;

Modules

capability_check

Capability and context-window guard for the routing / failover path.

context

RequestContext carries authenticated identity, trace IDs, and credentials through the request lifecycle. Adapters are stateless — every call gets a fresh context.

error

Error types returned by provider adapters. The core layer maps these to HTTP status codes and decides retry strategy — adapters do not retry.

messages

OpenAI-compatible request/response shapes. Canonical wire format across all providers — adapters translate to/from provider-native formats.

pricing

Pricing tables per model. Values are a manually-curated snapshot taken from provider pricing pages; they are NOT refreshed automatically. effective_at records when each rate took effect and lets us replay historical telemetry against the correct rate. To refresh rates, edit data/pricing.toml and append new entries — see scripts/refresh-pricing.sh for the manual workflow. See also docs/02-provider-adapter-guide.md.

provider

The Provider trait every adapter implements. See docs/02-provider-adapter-guide.md for the contract and the worked Anthropic example.

providers

Lightweight model → provider lookup.

url_guard

SSRF and header-injection guard for customer-supplied provider URLs.

usage

Token usage and cost accounting. Token counts come from provider responses — we never estimate locally for billing.