llmshim 0.3.5__tar.gz → 0.3.7__tar.gz

{llmshim-0.3.5 → llmshim-0.3.7}/CLAUDE.md

@@ -10,13 +10,14 @@ A pure Rust LLM API translation layer. Takes OpenAI-format JSON requests, transl
 
 This is a public crate on crates.io. Do NOT make breaking changes to `pub` items in `src/lib.rs`, `src/router.rs`, `src/provider.rs`, `src/error.rs`, `src/fallback.rs`, `src/log.rs`, `src/config.rs`, `src/models.rs`, or `src/vision.rs` without a semver bump.
 
-## Supported models
-
-- **OpenAI:** `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5`, `gpt-5.5-pro`, `gpt-5.4`, `gpt-5.4-pro`, `gpt-5.4-mini`, `gpt-5.4-nano`
-- **Anthropic:** `claude-opus-5`, `claude-opus-4-8`, `claude-sonnet-5`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-haiku-4-5-20251001`
-- **Gemini:** `gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-3.1-flash-lite`
-- **xAI:** `grok-4.6`, `grok-4.5`, `grok-4.3`, `grok-4.20-multi-agent-beta-0309`, `grok-4.20-beta-0309-reasoning`, `grok-4.20-beta-0309-non-reasoning`
-- **OpenRouter:** not enumerated (huge/dynamic catalog) — any `openrouter/<vendor>/<model>` slug routes through, e.g. `openrouter/anthropic/claude-sonnet-4.5`.
+## Advertised models
+
+- **OpenAI:** `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`
+- **ChatGPT subscription (OAuth):** only `chatgpt/gpt-6-astra`, `chatgpt/gpt-5.6-sol`, `chatgpt/gpt-5.6-terra`, and `chatgpt/gpt-5.6-luna`. `CHATGPT_MODELS` in `src/models.rs` is shared by discovery, CLI selection, and validation; older/unlisted models fail before authentication or network calls.
+- **Anthropic:** `claude-fable-5-1`, `claude-opus-5`, `claude-sonnet-5`, `claude-haiku-4-5-20251001`
+- **Gemini:** `gemini-3.8-flash`, `gemini-3.5-flash-lite`
+- **xAI:** `grok-4.6`
+- **OpenRouter:** not enumerated (huge/dynamic catalog) — any `openrouter/<vendor>/<model>` slug routes through, e.g. `openrouter/anthropic/claude-sonnet-5`.
 - **vLLM / SGLang:** not enumerated (self-hosted) — any `vllm/<served-model>` or `sglang/<served-model>` routes through to the configured server, e.g. `sglang/Qwen/Qwen3.6-35B-A3B-FP8`.
 
 ## Build & Test
@@ -36,10 +37,60 @@ API keys: `~/.llmshim/config.toml` (via `llmshim configure`) or env vars `OPENAI
 
 ## Architecture
 
+### Curated discovery
+
+`src/models.rs::MODELS` is the single advertised list, imported directly by
+`src/main.rs` and used by `available_models()` for proxy discovery. Keep the
+current model in each retained tier: four OpenAI, four Anthropic, two stable
+Gemini, one xAI, and four ChatGPT routes. Do not add previews or bring back
+superseded generations without an explicit catalog decision.
+
+Historical metadata lives in private `LEGACY_MODELS` and remains queryable via
+`spec()`. Pruning discovery does not delete legacy transforms, tests, or
+explicit-ID routing. ChatGPT keeps its separately authorized four-ID request
+allowlist. Keep reader-facing model tables and examples current; retain the
+actual model IDs in historical benchmark results and regression fixtures.
+
 ### Value-based transforms, no canonical struct
 
 Requests flow as `serde_json::Value`. Each provider's transform takes raw JSON and maps only what it understands. Provider-specific features use `x-anthropic`, `x-gemini`, `x-openrouter`, `x-vllm`, `x-sglang` namespaces.
 
+**ChatGPT OAuth (`src/providers/chatgpt/`).** Run `llmshim login chatgpt` for
+device-code authentication, `login chatgpt --status` for a local check, and
+`logout chatgpt` to remove the selected cache. The independent default cache
+is `~/.llmshim/chatgpt/auth.json`; `CHATGPT_TOKEN_DIR`/`CHATGPT_AUTH_FILE` can
+override it. Never read or overwrite Codex credentials implicitly. The
+object-safe `Provider::prepare_request` hook defaults to `transform_request`;
+ChatGPT uses it to refresh asynchronously with cross-process file locking and
+atomic owner-only token writes. Requests never initiate interactive login.
+
+The subscription backend requires SSE, `stream: true`, and `store: false`.
+ChatGPT reuses the Responses translator, enforces the backend field allowlist
+after `x-chatgpt` overrides, and aggregates a validated terminal event for
+non-streaming callers. EOF or `[DONE]` without a terminal event is an error.
+Preserve `chatgpt/<model>` in normalized responses and chunks: a bare GPT name
+is otherwise misattributed to API-key OpenAI by the proxy/gateway when both
+providers are registered. Astra preserves reasoning effort `max`; `none` and
+`minimal` clamp to `low`.
+For ChatGPT streaming, emit function calls from `response.output_item.done`
+with complete arguments. Forwarding `response.output_item.added` followed by
+argument-only deltas loses arguments at the proxy's `tool_call` boundary.
+Text and reasoning remain incremental.
+
+Offline coverage lives in `tests/unit_chatgpt.rs`. The live server check starts
+its own loopback CLI process and stops it on completion/failure:
+
+```bash
+cargo test --features proxy --test integration_chatgpt_proxy -- --ignored --nocapture
+```
+
+It checks all four models through `/v1/chat` and `/v1/chat/stream`, provider
+identity with API-key OpenAI also registered, model discovery, old-model
+rejection, Astra tool calls (normal and streaming), a tool-result round trip,
+and image input. It uses the saved ChatGPT login and consumes
+subscription usage; it is ignored during offline CI. Mount the whole token
+directory writable for container use so refresh locks and atomic saves work.
+
 **Self-hosted passthrough providers (vLLM / SGLang).** `src/providers/openai_compat.rs` is one generic OpenAI Chat Completions passthrough backing both `vllm` and `sglang` (`OpenAiCompatible::new(name, base_url, api_key: Option)`, registered per env base URL). Two things differ from the hosted providers: the **base URL is configuration** (local `http://localhost:8000/v1` vs remote `https://host/v1`), and **auth is optional** (self-hosted servers are unauthenticated unless launched with `--api-key`, so the `Authorization` header is sent only when a key is set). Passthrough transforms; `reasoning`/`reasoning_content` normalized to `reasoning_content` (vLLM is migrating the field name); `reasoning_effort` forwarded as-is (honored per-model, not clamped); server-specific params go under `x-<name>` (`chat_template_kwargs`, `separate_reasoning`, `guided_json`, `top_k`, …). Note: reasoning/tool parsing are **launch-time server flags** (`--reasoning-parser`, `--tool-call-parser`), so a request only gets that behavior if the server was started for it — llmshim can't enable it per request.
 
 **OpenRouter is the one passthrough provider.** Every other provider translates the OpenAI-format input *away* to a native dialect; OpenRouter (`src/providers/openrouter.rs`) *is* OpenAI Chat Completions, so its transforms are near-identity — messages, tools, vision (`image_url`), and `response_format` are forwarded unchanged; `reasoning_effort` maps 1:1 to OpenRouter's `reasoning:{effort}` (its effort vocabulary is a superset, so no clamping); `message.reasoning` is normalized to `reasoning_content` on responses. OpenRouter models are **not enumerated** in `src/models.rs` (the catalog is huge and dynamic) — any `openrouter/<vendor>/<model>` slug routes through. `x-openrouter` carries OpenRouter-only controls (`provider`, `models`, `transforms`, `route`, native `reasoning`; plus `http_referer`/`x_title` which become headers). The `middle-out` transform is disabled by default for faithful passthrough. Uses `image_url` (Chat Completions) vision via `vision::to_openai_chat`.
@@ -48,8 +99,8 @@ Requests flow as `serde_json::Value`. Each provider's transform takes raw JSON a
 
 ```
 llmshim::completion(router, request)
-  → router.resolve("anthropic/claude-sonnet-4-6")   // parse "provider/model"
-  → provider.transform_request(model, &value)        // OpenAI JSON → provider-native
+  → router.resolve("anthropic/claude-sonnet-5")   // parse "provider/model"
+  → provider.prepare_request(model, &value).await    // refresh OAuth if needed, then transform
   → client.send(provider_request)                    // HTTP
   → provider.transform_response(model, body)         // provider-native → OpenAI JSON
 ```
@@ -58,12 +109,23 @@ llmshim::completion(router, request)
 
 Every provider implements: `transform_request`, `transform_response`, `transform_stream_chunk`.
 
+Non-streaming OpenAI/xAI Responses, Gemini and Anthropic transformations require
+supported terminal metadata. Do not turn absent or unrecognized status into
+`finish_reason: "stop"`. Rejected metadata returns a fixed 502 provider error
+without copying provider-supplied status/content into the diagnostic. Tests
+live in `tests/support/completion_status.rs`, included by `unit_openai`.
+Streaming status handling is separate and is not changed by this policy.
+
 ### Router (`src/router.rs`)
 
-Parses `"provider/model"` strings by splitting on the **first** `/` only, so an OpenRouter slug's internal slash survives (`openrouter/anthropic/claude-sonnet-4.5` → provider `openrouter`, model `anthropic/claude-sonnet-4.5`). Auto-infers provider from prefix (`gpt*`/`o*` → openai, `claude*` → anthropic, `gemini*` → gemini, `grok*` → xai); **OpenRouter, vLLM, and SGLang have no prefix inference** — their slugs collide with everyone's, so address them explicitly (`openrouter/…`, `vllm/…`, `sglang/…`); the first-slash split also preserves HF-style served-model slugs (`vllm/meta-llama/Llama-3.1-8B-Instruct`). Supports aliases. `Router::from_env()` reads API-key env vars, plus `VLLM_BASE_URL` / `SGLANG_BASE_URL` (+ optional `*_API_KEY`) for the self-hosted providers.
+Parses `"provider/model"` strings by splitting on the **first** `/` only, so an OpenRouter slug's internal slash survives (`openrouter/anthropic/claude-sonnet-5` → provider `openrouter`, model `anthropic/claude-sonnet-5`). Auto-infers provider from prefix (`gpt*`/`o*` → openai, `claude*` → anthropic, `gemini*` → gemini, `grok*` → xai); **OpenRouter, vLLM, and SGLang have no prefix inference** — their slugs collide with everyone's, so address them explicitly (`openrouter/…`, `vllm/…`, `sglang/…`); the first-slash split also preserves HF-style served-model slugs (`vllm/meta-llama/Llama-3.1-8B-Instruct`). Supports aliases. `Router::from_env()` reads API-key env vars, plus `VLLM_BASE_URL` / `SGLANG_BASE_URL` (+ optional `*_API_KEY`) for the self-hosted providers.
 
 ### HTTP Client (`src/client.rs`)
 
+Automatic redirects are disabled on the shared client. Keep prompts and
+provider-specific credential headers at the configured endpoint; 3xx responses
+remain provider errors. Callers must configure the final URL directly.
+
 `ShimClient` with shared connection pool (`LazyLock`), HTTP/2, gzip/brotli/zstd compression, TCP keepalive + nodelay. Automatic retry (3 attempts by default) on transport errors and 429/500/502/503/504/529 status codes. This is the **reactive** layer: on a retryable *response* it honors the server's `Retry-After` header (integer seconds or HTTP-date) and provider reset hints (OpenAI `x-ratelimit-reset-*`, Anthropic `anthropic-ratelimit-*-reset`), clamped to a cap and nudged with a little jitter; when there's no server hint (or a transport error) it falls back to full-jitter exponential backoff (uniform in `[0, min(cap, base·2^attempt)]`) to avoid a thundering herd. Tunable via `LLMSHIM_MAX_RETRIES` and `LLMSHIM_MAX_BACKOFF_SECS`. `warmup()` pre-establishes TCP+TLS connections. `SseStream` buffers bytes, extracts `data:` lines, routes through provider's `transform_stream_chunk`.
 
 ### Fallback chains (`src/fallback.rs`)
@@ -87,7 +149,36 @@ Callers pass provider-specific controls under these keys. Each provider copies w
 
 ### Unified reasoning controls
 
-Two knobs work across every provider: `reasoning_effort` (`none|low|medium|high|xhigh|max`) and `reasoning_mode` (`standard|pro`). A third, `reasoning_summary` (`auto|none`), controls reasoning-text visibility → Anthropic `thinking.display` (`auto`→`summarized`, the default when `reasoning_effort` is present so newer models like Sonnet 5 / Opus 4.7-4.8 return reasoning text instead of the API-default `omitted`; `none`→`omitted` for lower latency). Applies to both the adaptive and pre-4.6 enabled thinking builders; a caller-supplied `thinking` block bypasses it. Each provider transform maps them to its native dialect, **clamping to the nearest tier the target model accepts** (all boundaries verified live — e.g. `max` is native only on OpenAI gpt-5.6; Anthropic 4.6 rejects `xhigh` but has `max`; Gemini's enum tops out at `high`; xAI grok-4.20 models reject any reasoning param). `mode: "pro"` is native on OpenAI gpt-5.6/-pro models (`reasoning.mode`), emulated as a one-tier effort bump elsewhere; explicit `none` always wins. Native passthrough (`x-openai.reasoning`, `x-anthropic.thinking`, `x-gemini.thinkingConfig`) bypasses the mapping entirely and always takes precedence. **Full per-provider mapping tables: `docs/src/guides/reasoning.md`** — update it and the pinning tests in `tests/unit_*.rs` together whenever a mapping changes.
+**Fable compatibility (verified September 2026).** Advertise Fable 5.1;
+retain Fable 5 behavior and metadata for explicit requests. Both use always-on adaptive thinking; unified `none` clamps to `low`,
+and native disabled/manual thinking fails locally. Strip `temperature`,
+`top_p`, and `top_k` for Fable and Opus 5 even without an explicit thinking
+object. Fable 5.1 rejects forced tool selection (`required` / `any` / `tool`);
+do not silently turn it into `auto`. Fable 5 still accepts forced tools.
+Both versions reject assistant prefill. Refusal stop reasons on Fable/Opus 5
+map to `content_filter` in normal and streaming engine responses.
+
+Fable 5.1's thinking signatures are conversation-bound. Preserve appended
+system turns instead of hoisting them into the initial prompt. Keep the
+initial system/tools/message prefix stable during a signed-thinking round
+trip; errors must not become success. Test with the
+`thinking-binding-controls-2026-08-01` beta and
+`thinking.block_binding.prefix_mismatch_behavior: "error"` so older API
+accounts exercise enforcement too. The provider handles incompatible
+thinking on a switch to older models; do not infer signatures from text.
+
+`tests/unit_fable.rs` pins these rules. Live checks for Fable 5, Fable 5.1,
+Opus 5, Gemini 3.8 Flash, and Grok 4.6:
+
+```bash
+cargo test --features proxy --test integration_current_models -- --ignored --nocapture
+```
+
+The live tests consume API usage and are ignored during offline preflight.
+Gemini 3.8 Flash and Opus 5 already had catalog/adapter support; Grok 4.6 is
+the verified xAI model ID. Keep the ChatGPT four-model allowlist independent.
+
+Two knobs work across every provider: `reasoning_effort` (`none|low|medium|high|xhigh|max`) and `reasoning_mode` (`standard|pro`). A third, `reasoning_summary` (`auto|none`), controls reasoning-text visibility → Anthropic `thinking.display` (`auto`→`summarized`, the default when `reasoning_effort` is present so newer models like Sonnet 5 / Opus 4.7-4.8 return reasoning text instead of the API-default `omitted`; `none`→`omitted` for lower latency). Applies to both the adaptive and pre-4.6 enabled thinking builders; a caller-supplied `thinking` block bypasses it. Each provider transform maps them to its native dialect, **clamping to the nearest tier the target model accepts** (all boundaries verified live — e.g. `max` is native on OpenAI gpt-5.6 and GPT-6 Astra; Anthropic 4.6 rejects `xhigh` but has `max`; Gemini's enum tops out at `high`; xAI grok-4.20 models reject any reasoning param). `mode: "pro"` is native on OpenAI gpt-5.6/-pro models (`reasoning.mode`), emulated as a one-tier effort bump elsewhere; explicit `none` always wins. Native passthrough (`x-openai.reasoning`, `x-anthropic.thinking`, `x-gemini.thinkingConfig`) bypasses the mapping entirely and always takes precedence. **Full per-provider mapping tables: `docs/src/guides/reasoning.md`** — update it and the pinning tests in `tests/unit_*.rs` together whenever a mapping changes.
 
 ### Tool format translation
 

{llmshim-0.3.5 → llmshim-0.3.7}/Cargo.lock

@@ -371,7 +371,7 @@ dependencies = [
  "crossterm_winapi",
  "mio",
  "parking_lot",
- "rustix",
+ "rustix 0.38.44",
  "signal-hook",
  "signal-hook-mio",
  "winapi",
@@ -503,6 +503,16 @@ dependencies = [
  "percent-encoding",
 ]
 
+[[package]]
+name = "fs2"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9564fc758e15025b46aa6643b1b77d047d1a56a1aea6e01002ac0c7026876213"
+dependencies = [
+ "libc",
+ "winapi",
+]
+
 [[package]]
 name = "futures"
 version = "0.3.32"
@@ -950,6 +960,12 @@ version = "0.4.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d26c52dbd32dccf2d10cac7725f8eae5296885fb5703b261f7d0a0739ec807ab"
 
+[[package]]
+name = "linux-raw-sys"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32a66949e030da00e8c7d4434b251670a91556f4144941d37452769c25d58a53"
+
 [[package]]
 name = "litemap"
 version = "0.8.1"
@@ -958,22 +974,25 @@ checksum = "6373607a59f0be73a39b6fe456b8192fcc3585f602af20751600e974dd455e77"
 
 [[package]]
 name = "llmshim"
-version = "0.3.5"
+version = "0.3.7"
 dependencies = [
  "async-stream",
  "async-trait",
  "axum",
+ "base64",
  "bytes",
  "chrono",
  "crossterm",
  "dirs",
  "eventsource-stream",
+ "fs2",
  "futures",
  "mockito",
  "redis",
  "reqwest",
  "serde",
  "serde_json",
+ "tempfile",
  "thiserror",
  "tokio",
  "tokio-stream",
@@ -1441,10 +1460,23 @@ dependencies = [
  "bitflags",
  "errno",
  "libc",
- "linux-raw-sys",
+ "linux-raw-sys 0.4.15",
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "rustix"
+version = "1.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6fe4565b9518b83ef4f91bb47ce29620ca828bd32cb7e408f0062e9930ba190"
+dependencies = [
+ "bitflags",
+ "errno",
+ "libc",
+ "linux-raw-sys 0.12.1",
+ "windows-sys 0.61.2",
+]
+
 [[package]]
 name = "rustls"
 version = "0.23.37"
@@ -1737,6 +1769,19 @@ dependencies = [
  "syn",
 ]
 
+[[package]]
+name = "tempfile"
+version = "3.27.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32497e9a4c7b38532efcdebeef879707aa9f794296a4f0244f6f69e9bc8574bd"
+dependencies = [
+ "fastrand",
+ "getrandom 0.3.4",
+ "once_cell",
+ "rustix 1.1.4",
+ "windows-sys 0.61.2",
+]
+
 [[package]]
 name = "thiserror"
 version = "2.0.18"

{llmshim-0.3.5 → llmshim-0.3.7}/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "llmshim"
-version = "0.3.5"
+version = "0.3.7"
 edition = "2021"
 description = "Blazing fast LLM API translation layer in pure Rust"
 license = "MIT OR Apache-2.0"
@@ -36,6 +36,9 @@ chrono = { version = "0.4", features = ["serde"] }
 crossterm = "0.28"
 toml = "0.8"
 dirs = "6"
+base64 = "0.22"
+tempfile = "3"
+fs2 = "0.4"
 
 # Proxy dependencies (feature-gated)
 axum = { version = "0.8", features = ["json"], optional = true }

{llmshim-0.3.5 → llmshim-0.3.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llmshim
-Version: 0.3.5
+Version: 0.3.7
 Classifier: Development Status :: 4 - Beta
 Classifier: Intended Audience :: Developers
 Classifier: License :: OSI Approved :: MIT License
@@ -66,7 +66,7 @@ Then address them via the model string — `vllm/<served-model>` or
 ```python
 import llmshim
 
-resp = llmshim.chat("claude-sonnet-4-6", "What is Rust?")
+resp = llmshim.chat("claude-sonnet-5", "What is Rust?")
 print(resp["message"]["content"])
 ```
 
@@ -74,7 +74,7 @@ With options (all map to the API's provider-agnostic `config`):
 
 ```python
 resp = llmshim.chat(
-    "openai/gpt-5.5",
+    "openai/gpt-5.6-sol",
     "Explain quicksort",
     max_tokens=500,
     temperature=0.7,
@@ -88,7 +88,7 @@ resp = llmshim.chat(
 With message history:
 
 ```python
-resp = llmshim.chat("claude-sonnet-4-6", [
+resp = llmshim.chat("claude-sonnet-5", [
     {"role": "system", "content": "You are a pirate."},
     {"role": "user", "content": "Hello!"},
 ], max_tokens=500)
@@ -97,7 +97,7 @@ resp = llmshim.chat("claude-sonnet-4-6", [
 ## Streaming
 
 ```python
-for event in llmshim.stream("claude-sonnet-4-6", "Write a poem"):
+for event in llmshim.stream("claude-sonnet-5", "Write a poem"):
     if event["type"] == "content":
         print(event["text"], end="", flush=True)
     elif event["type"] == "reasoning":
@@ -113,13 +113,13 @@ Switch models mid-conversation. History carries over.
 ```python
 messages = [{"role": "user", "content": "What is a closure?"}]
 
-r1 = llmshim.chat("claude-sonnet-4-6", messages, max_tokens=500)
+r1 = llmshim.chat("claude-sonnet-5", messages, max_tokens=500)
 print(f"Claude: {r1['message']['content']}")
 
 messages.append({"role": "assistant", "content": r1["message"]["content"]})
 messages.append({"role": "user", "content": "Now explain differently."})
 
-r2 = llmshim.chat("gpt-5.5", messages, max_tokens=500)
+r2 = llmshim.chat("gpt-5.6-sol", messages, max_tokens=500)
 print(f"GPT: {r2['message']['content']}")
 ```
 
@@ -147,7 +147,7 @@ print(resp["message"]["content"])  # answer
 
 For full native control, bypass the unified mapping with a namespaced
 `provider_config` (see below), e.g.
-`provider_config={"x-anthropic": {"thinking": {"type": "enabled", "budget_tokens": 4000}}}`.
+`provider_config={"x-anthropic": {"thinking": {"type": "adaptive"}, "output_config": {"effort": "high"}}}`.
 
 ## Provider-Specific Controls (`provider_config`)
 
@@ -163,10 +163,8 @@ resp = llmshim.chat(
     "Solve this step by step: 17 * 23",
     max_tokens=4000,
     provider_config={
-        # native Anthropic extended-thinking control
-        "x-anthropic": {"thinking": {"type": "enabled", "budget_tokens": 4000}},
-        # structured output
-        "response_format": {"type": "json_object"},
+        # native Anthropic adaptive-thinking control
+        "x-anthropic": {"thinking": {"type": "adaptive"}, "output_config": {"effort": "high"}},
     },
 )
 ```
@@ -175,7 +173,7 @@ OpenRouter routing preferences use the `x-openrouter` namespace:
 
 ```python
 resp = llmshim.chat(
-    "openrouter/anthropic/claude-sonnet-4.5",
+    "openrouter/anthropic/claude-sonnet-5",
     "Hello",
     max_tokens=200,
     provider_config={"x-openrouter": {"provider": {"sort": "throughput"}}},
@@ -198,7 +196,7 @@ tools = [{
     },
 }]
 
-resp = llmshim.chat("claude-sonnet-4-6", "Weather in Tokyo?", max_tokens=500, tools=tools)
+resp = llmshim.chat("claude-sonnet-5", "Weather in Tokyo?", max_tokens=500, tools=tools)
 for tc in resp["message"].get("tool_calls", []):
     print(f"{tc['function']['name']}({tc['function']['arguments']})")
 ```
@@ -209,10 +207,10 @@ Tools are accepted in OpenAI Chat Completions format and auto-translated to each
 
 ```python
 resp = llmshim.chat(
-    "anthropic/claude-sonnet-4-6",
+    "anthropic/claude-sonnet-5",
     "Hello",
     max_tokens=100,
-    fallback=["openai/gpt-5.6-sol", "gemini/gemini-3.5-flash"],
+    fallback=["openai/gpt-5.6-sol", "gemini/gemini-3.8-flash"],
 )
 ```
 
@@ -242,7 +240,7 @@ common ones are re-exported at the top level) for static type-checking:
 ```python
 from llmshim.types import ChatResponse, StreamEvent, Message, Config
 
-resp: ChatResponse = llmshim.chat("claude-sonnet-4-6", "hi")
+resp: ChatResponse = llmshim.chat("claude-sonnet-5", "hi")
 ```
 
 Available: `ChatRequest`, `ChatResponse`, `Config`, `Message`, `ToolCall`,
@@ -281,14 +279,14 @@ pytest tests/
 `test_e2e.py` is a separate LIVE suite that spawns the real binary and makes
 billed provider calls; run it only when you deliberately want to hit real APIs.
 
-## Supported Models
+## Advertised Models
 
 | Provider | Models |
 |----------|--------|
-| OpenAI | `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5`, `gpt-5.5-pro`, `gpt-5.4`, `gpt-5.4-pro`, `gpt-5.4-mini`, `gpt-5.4-nano` |
-| Anthropic | `claude-opus-5`, `claude-opus-4-8`, `claude-sonnet-5`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-haiku-4-5-20251001` |
-| Gemini | `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-3.1-flash-lite` |
-| xAI | `grok-4.6`, `grok-4.5`, `grok-4.3`, `grok-4.20-multi-agent-beta-0309`, `grok-4.20-beta-0309-reasoning`, `grok-4.20-beta-0309-non-reasoning` |
+| OpenAI | `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna` |
+| Anthropic | `claude-fable-5-1`, `claude-opus-5`, `claude-sonnet-5`, `claude-haiku-4-5-20251001` |
+| Gemini | `gemini-3.8-flash`, `gemini-3.5-flash-lite` |
+| xAI | `grok-4.6` |
 
 Call `llmshim.models()` for the live list filtered to your configured providers.
 
@@ -299,7 +297,7 @@ any model the upstream serves is reachable, so they aren't in the table above.
 
 | Provider | Address as | Env vars | Native controls |
 |----------|-----------|----------|-----------------|
-| OpenRouter | `openrouter/<vendor>/<model>` (e.g. `openrouter/anthropic/claude-sonnet-4.5`) | `OPENROUTER_API_KEY` (or `llmshim.configure(openrouter=...)`) | `provider_config={"x-openrouter": {...}}` (`provider`, `models`, `transforms`) |
+| OpenRouter | `openrouter/<vendor>/<model>` (e.g. `openrouter/anthropic/claude-sonnet-5`) | `OPENROUTER_API_KEY` (or `llmshim.configure(openrouter=...)`) | `provider_config={"x-openrouter": {...}}` (`provider`, `models`, `transforms`) |
 | vLLM | `vllm/<served-model>` | `VLLM_BASE_URL` (+ optional `VLLM_API_KEY`) | `provider_config={"x-vllm": {...}}` |
 | SGLang | `sglang/<served-model>` | `SGLANG_BASE_URL` (+ optional `SGLANG_API_KEY`) | `provider_config={"x-sglang": {...}}` |
 

{llmshim-0.3.5 → llmshim-0.3.7}/README.md

@@ -1,6 +1,6 @@
 # llmshim
 
-A blazing-fast LLM API translation layer written in **pure Rust**. One request format, every provider — OpenAI, Anthropic, Google Gemini, xAI, OpenRouter, and self-hosted vLLM / SGLang.
+A blazing-fast LLM API translation layer written in **pure Rust**. One request format, every provider — OpenAI, ChatGPT subscriptions, Anthropic, Google Gemini, xAI, OpenRouter, and self-hosted vLLM / SGLang.
 
 Send an OpenAI-style request, pick any model, and llmshim translates it to that provider's native API (and translates the response back). Switch providers by changing one string.
 
@@ -51,7 +51,7 @@ export OPENROUTER_API_KEY=sk-or-...
 ```
 
 Reach any model through [OpenRouter](https://openrouter.ai) by addressing it as
-`openrouter/<vendor>/<model>` (e.g. `openrouter/anthropic/claude-sonnet-4.5`).
+`openrouter/<vendor>/<model>` (e.g. `openrouter/anthropic/claude-sonnet-5`).
 OpenRouter is OpenAI Chat Completions-compatible, so tools, vision, streaming,
 and `reasoning_effort` all pass through; OpenRouter-only controls (provider
 routing, model fallbacks, transforms) go under an `x-openrouter` key.
@@ -76,8 +76,71 @@ llmshim configure          # interactive prompt
 
 ---
 
+## ChatGPT subscription (OAuth)
+
+Sign in once with a ChatGPT account, then use `chatgpt/<model>` from Rust,
+the CLI, or any proxy client. No `OPENAI_API_KEY` is needed for this route.
+The device-code flow follows [LiteLLM's ChatGPT provider](https://docs.litellm.ai/docs/providers/chatgpt).
+
+```bash
+llmshim login chatgpt             # open the printed URL and enter the code
+llmshim login chatgpt --status    # inspect the saved login without network access
+llmshim chat                     # select a ChatGPT model
+# or: llmshim proxy
+```
+
+If needed, enable device-code login in your ChatGPT security settings or
+workspace permissions ([OpenAI authentication docs](https://learn.chatgpt.com/docs/auth#preferred-device-code-authentication-beta)).
+
+```json
+{
+  "model": "chatgpt/gpt-6-astra",
+  "messages": [{"role": "user", "content": "Hello!"}]
+}
+```
+
+Tokens live in `~/.llmshim/chatgpt/auth.json`; expired access tokens refresh
+automatically, with file locking and atomic saves. `CHATGPT_TOKEN_DIR` and
+`CHATGPT_AUTH_FILE` select a different cache (LiteLLM's flat auth-file format
+is supported). This cache is independent of Codex's login. Run
+`llmshim logout chatgpt` to remove the selected local cache; this does not
+revoke the session at OpenAI. Login is explicit, never started inside a proxy
+request. Restart an existing proxy after the first login so it registers the
+provider. For containers, mount the token directory writable and set
+`CHATGPT_TOKEN_DIR` to its container path.
+
+Both completion and streaming calls use the subscription Responses backend.
+Non-streaming calls collect the upstream stream into one normal response.
+Tools, images, and reasoning use the existing Responses translation;
+`x-chatgpt` supplies supported native fields (under `provider_config` in proxy
+requests). The backend requires `store: false` and `stream: true` and rejects
+token limits, sampling fields, and metadata, so these constraints also apply
+to native overrides. Bare `gpt-*` names still route to API-key OpenAI.
+The ChatGPT route supports only `chatgpt/gpt-6-astra`, `chatgpt/gpt-5.6-sol`,
+`chatgpt/gpt-5.6-terra`, and `chatgpt/gpt-5.6-luna`. Older and unlisted model
+IDs return a local error before authentication or an upstream request.
+Access to these models and usage limits depend on the ChatGPT account.
+
+See [provider configuration](docs/src/reference/configuration.md#chatgpt-oauth)
+for endpoint and header overrides.
+
+## Endpoint redirects
+
+The shared HTTP client does not follow redirects. Configure the final API URL
+directly: a 3xx response is returned as a provider error instead of forwarding
+the prompt and provider-specific credential headers to another endpoint.
+This applies to both streaming and non-streaming requests.
+
 ## Use it from Rust
 
+Non-streaming OpenAI Responses, xAI Responses, Gemini, and Anthropic results
+require supported terminal metadata. Missing, malformed, or unsupported
+statuses return a `ProviderError` (502) with a fixed diagnostic instead of
+being interpreted as successful completion. Known completion, output-limit,
+filtering and Anthropic tool-call mappings remain available. Additional native
+reasons require an explicit mapping; streaming transforms are unchanged.
+Provider mocks should include the native terminal status/stop reason.
+
 ```bash
 cargo add llmshim tokio serde_json
 ```
@@ -141,6 +204,7 @@ cargo install llmshim --features proxy    # from source (any platform)
 llmshim                     # show help
 llmshim chat                # interactive multi-model chat (streaming, /model to switch)
 llmshim configure           # set API keys
+llmshim login chatgpt        # sign in with a ChatGPT subscription
 llmshim set <key> <value>   # set a config value
 llmshim list                # show configured keys
 llmshim models              # list available models
@@ -298,7 +362,7 @@ resp = llmshim.chat(
     "anthropic/claude-sonnet-5",
     "Hello",
     max_tokens=100,
-    fallback=["openai/gpt-5.6-sol", "gemini/gemini-3.5-flash"],
+    fallback=["openai/gpt-5.6-sol", "gemini/gemini-3.8-flash"],
 )
 ```
 
@@ -358,16 +422,22 @@ Standard library only. Full docs: [`clients/ruby/README.md`](clients/ruby/README
 
 ---
 
-## Supported models
+## Advertised models
 
 | Provider | Models | Reasoning visible |
 |----------|--------|-------------------|
-| **OpenAI** | `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5`, `gpt-5.5-pro`, `gpt-5.4`, `gpt-5.4-pro`, `gpt-5.4-mini`, `gpt-5.4-nano` | Yes (summaries) |
-| **Anthropic** | `claude-opus-5`, `claude-opus-4-8`, `claude-sonnet-5`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-haiku-4-5-20251001` | Yes (full thinking) |
-| **Google Gemini** | `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3.5-flash-lite`, `gemini-3.1-flash-lite` | Yes (thought summaries) |
-| **xAI** | `grok-4.6`, `grok-4.5`, `grok-4.3`, `grok-4.20-multi-agent-beta-0309`, `grok-4.20-beta-0309-reasoning`, `grok-4.20-beta-0309-non-reasoning` | No (hidden) |
-
-Use a bare model name (auto-detected by prefix) or an explicit `provider/model` string.
+| **OpenAI** | `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna` | Yes (summaries) |
+| **Anthropic** | `claude-fable-5-1`, `claude-opus-5`, `claude-sonnet-5`, `claude-haiku-4-5-20251001` | Yes (thinking summaries) |
+| **Google Gemini** | `gemini-3.8-flash`, `gemini-3.5-flash-lite` | Yes (thought summaries) |
+| **xAI** | `grok-4.6` | No (hidden) |
+
+The CLI and server advertise these current tiers. ChatGPT subscription access
+uses the same four OpenAI models under `chatgpt/`. OpenRouter and self-hosted
+providers accept caller-selected IDs without a fixed advertised list.
+
+Use a bare model name (auto-detected by prefix) or an explicit `provider/model`
+string. Older explicit IDs retain their provider routing and metadata; the
+ChatGPT route continues to enforce its four-model allowlist.
 
 ## Docker
 

{llmshim-0.3.5 → llmshim-0.3.7}/docs/src/concepts/conversations.md

@@ -16,7 +16,7 @@ Start with a history:
 Send it to another provider by keeping `messages` and changing the address:
 
 ```diff
-- "model": "anthropic/claude-opus-4-8"
+- "model": "anthropic/claude-opus-5"
 + "model": "openai/gpt-5.6-sol"
 ```
 

{llmshim-0.3.5 → llmshim-0.3.7}/docs/src/concepts/routing.md

@@ -10,9 +10,9 @@ The most explicit form is `provider/model`:
 
 ```text
 openai/gpt-5.6-sol
-anthropic/claude-opus-4-8
-gemini/gemini-3.5-flash
-xai/grok-4.5
+anthropic/claude-opus-5
+gemini/gemini-3.8-flash
+xai/grok-4.6
 ```
 
 The part before the first slash is the Router registration key. The remainder
@@ -39,11 +39,16 @@ explicit address when inference cannot identify the provider.
 Resolution succeeds only when the selected provider key is registered on the
 Router. The built-in `Router::from_env()` registers OpenAI, Anthropic, Gemini,
 and xAI only when their corresponding environment variables are present.
+ChatGPT is registered when its OAuth cache exists; sign in with
+`llmshim login chatgpt` before starting the proxy. Use `chatgpt/<model>` to
+select subscription access. Bare GPT names continue to use OpenAI API keys.
 
 The static model registry powers `llmshim models` and `GET /v1/models`. Those
 commands are discovery aids, filtered to configured providers. The registry is
-not an allowlist: routing does not reject a model merely because it is absent
-from that list.
+not generally an allowlist: most providers accept models absent from that list.
+ChatGPT accepts only `chatgpt/gpt-6-astra` and the three
+`chatgpt/gpt-5.6-{sol,terra,luna}` models. The provider rejects other IDs
+before authentication or network calls, including through aliases.
 
 For that reason, this documentation does not maintain another static model
 table. Use runtime discovery for the current curated list.
@@ -54,7 +59,7 @@ Rust applications can attach a one-level alias while building a Router:
 
 ```rust
 let router = llmshim::router::Router::from_env()
-    .alias("smart", "anthropic/claude-opus-4-8");
+    .alias("smart", "anthropic/claude-opus-5");
 ```
 
 The Router checks an alias before parsing the provider address. An alias target
@@ -67,9 +72,10 @@ API, or language clients.
 
 ## Environment variables versus `config.toml`
 
-`Router::from_env()` means exactly what its name says: it reads
+`Router::from_env()` reads provider environment variables such as
 `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `GEMINI_API_KEY`, and `XAI_API_KEY`. It
-does not read `~/.llmshim/config.toml` by itself.
+does not read `~/.llmshim/config.toml` by itself. It also discovers the selected
+ChatGPT OAuth cache, whose default location is `~/.llmshim/chatgpt/auth.json`.
 
 The CLI and proxy call `llmshim::env::load_all()` before constructing their
 Router. That function loads the config file and fills only environment
@@ -84,4 +90,3 @@ let router = llmshim::router::Router::from_env();
 
 Applications that manage secrets themselves can call `Router::from_env()`
 directly or construct a Router by registering provider implementations.
-