Support matrix · formats and harnesses

What works with Torana.

Torana only sends recognized inference endpoints through its canonical IR and plugin pipeline. Other provider endpoints remain transparent reverse-proxy traffic. “Preserved” below means executable fixtures cover the wire behavior; it does not claim that every model or account enables the provider feature.

Provider formats

Format	Inference endpoints	Streaming	Tools	Prompt-cache behavior	Known boundary
OpenAI	`POST …/chat/completions` `POST …/responses`	SSE	Preserved and mutable under explicit grants	`prompt_cache_key`, `prompt_cache_retention`, and usage preserved; Responses compaction is opt-in	Provider-owned history behind `previous_response_id` is not rewritten
Anthropic	`POST …/messages`	SSE	Structured use/results and signatures preserved	Ordered `cache_control` breakpoints preserved	Both string and block-array system prompts are accepted
Gemini / Vertex AI	`POST …:generateContent` `POST …:streamGenerateContent`	SSE	Function calls/results, metadata, and signatures preserved	`cachedContent` references and usage preserved; Torana does not manage cache resources	Caller-supplied API key or bearer auth; ADC/service-account minting is not built in
Gemini Code Assist	`POST …:generateContent` `POST …:streamGenerateContent`	SSE	Function calls/results and signed parts preserved	Provider cache references and usage preserved	Clients without a base-URL setting require Torana’s optional, explicitly configured MITM ingress

Selective interception

Model-list, token-counting, account, quota, status, telemetry, update, MCP, and unknown auxiliary paths do not enter IR and do not run request or response plugins. Real-proxy tests cover auxiliary pass-through for all four formats, including method, path, query, body, permitted headers, upstream status, and response bytes.

Coding harnesses

Harness path	Torana format	Automated contract	Manual status
Claude Code via `ANTHROPIC_BASE_URL`	Anthropic Messages	System prompts, streaming tool use/results, cache markers, signatures, and auxiliary pass-through	Credentialed normal/tool/resume smoke remains a release gate
Codex custom provider	OpenAI Responses	Responses layout, opaque input items, streaming tools, and model-list pass-through	Credentialed normal/tool/resume smoke remains a release gate
OpenCode / Aider base URL	OpenAI or Anthropic	The selected provider wire has automated coverage	Credentialed client smoke remains a release gate
Antigravity / Code Assist	Gemini Code Assist	Wrapper data, signed parts, cache facts, stream framing, and auxiliary pass-through	Credentialed client smoke remains a release gate

Plugin SDKs

Go	ABI v1 SDK, helpers, test harness, official plugins, and host conformance.
Rust	ABI v1 bindings and safe wrappers. A clean Rust guest has built, installed, been digest-approved, and executed through the real host path. `v0.3.0` is tagged and published as a Go module; the crate is not yet on crates.io, so a Rust guest builds from the tag.
TypeScript / AssemblyScript	No supported SDK today. WASI alone is insufficient: a guest must also implement Torana’s protobuf ABI, allocator, framed host calls, and hook contract.
Other WASI languages	Possible through the documented raw ABI, but not advertised as supported until a maintained SDK and host-conformance guest exist.

How to read this matrix

Coverage is endpoint-specific: matching a provider name alone is not enough. Use the format and harness rows above to choose your route and distinguish automated wire tests from credentialed client checks.

Implementation evidence: harness boundary ↗ · prompt caching ↗ · benchmarks ↗