* fix: add kimi/moonshot to Anthropic tool thinking history normalizer
Kimi's Anthropic-compatible endpoint requires reasoning_content in
assistant tool call messages when thinking is enabled, same as DeepSeek.
The is_reasoning_content_compatible_identifier already included kimi/moonshot
but is_anthropic_tool_thinking_history_identifier was missing them, causing
400 errors on multi-turn conversations with tool calls.
Closes#3351
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* refactor: unify reasoning vendor hints into a single SSOT list
Replace the two separately-maintained vendor identifier lists with one
shared REASONING_VENDOR_HINTS constant and a single
is_reasoning_vendor_identifier predicate. The Anthropic
tool-thinking-history matcher and the openai_chat reasoning_content
matcher previously kept independent lists, which is exactly how
kimi/moonshot ended up in one but not the other (#3351). A single source
of truth keeps the two from drifting apart again.
Also add a Kimi regression test for the Anthropic tool-history
normalization path.
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Jason <farion1231@gmail.com>
When forward_with_retry fails for Codex endpoints (/chat/completions, /responses, /responses/compact), the handlers previously returned the bare ProxyError, surfacing a terse body to the client. Build a richer JSON error instead: the message embeds provider, model, endpoint and upstream status, and the error object carries those as structured fields alongside a stable cc_switch_* code.
Align map_proxy_error_to_status with ProxyError::into_response so the manually built responses use the canonical status codes (AuthError->401, Config/InvalidRequest->400, TransformError->422, StreamIdleTimeout->504, AlreadyRunning->409, NotRunning->503); previously these forwarder-level errors collapsed to 500.
Bump the default Opus route/model from claude-opus-4-7 to claude-opus-4-8
across provider presets (claude, claudeDesktop, hermes, openclaw, opencode,
universal), i18n locales (zh/en/ja/zh-TW), pricing seed data, and the
user-manual docs.
- Add claude-opus-4-8 pricing row ($5/$25/$0.50/$6.25); keep the 4-7 row
for historical usage stats (seeded via INSERT OR IGNORE).
- Claude Desktop proxy: accept bidirectional opus 4-7 <-> 4-8 route alias
during rollout so previously saved routes keep resolving.
- thinking_optimizer: route opus-4-8 through adaptive thinking and normalize
dotted model ids (also fixes dotted 4-6/4-7 falling back to legacy).
- usage_stats: normalize Bedrock/Vertex/aggregator opus-4-8 ids to base
pricing.
Also merge role:"system" messages into the Gemini systemInstruction in the
Anthropic->Gemini transform.
The usage dashboard previously only refreshed on app restart for users
who don't route through the cc-switch proxy. Two issues were involved:
1. The session-sync background task panicked when a Codex model name
contained non-ASCII characters (e.g. `【官】glm-5.1`), because
`normalize_codex_model` sliced `&name[name.len() - 11..]` without
verifying char boundaries. Once the task panicked, no session logs
were imported until the app was restarted (where startup-time
`rollup_and_prune` happened to flush pending data).
2. Even with sync working, the dashboard only polled every 30s and
skipped polling when the window was unfocused, so freshly-imported
data was invisible until the next poll or window refocus.
Fixes
-----
* `normalize_codex_model`: guard the 11-byte ISO-date suffix slice with
`is_char_boundary` + `is_ascii` checks. ASCII-only suffix means the
date-stripping logic is correct, and non-ASCII names (which can never
be valid date suffixes anyway) now bypass the slice safely.
* New `usage_events` module that emits `usage-log-recorded` to the
frontend whenever `proxy_request_logs` actually gains a new row.
Sources covered: proxy `log_request`, Claude/Codex/Gemini session
sync, and startup `rollup_and_prune`. Notifications use a global
`OnceLock<AppHandle>` so call sites that don't already hold an
`AppHandle` (e.g. `UsageLogger`) can notify without signature churn.
* 200ms debounce in `notify_log_recorded` collapses bursts (a single
Codex sync importing 3000+ entries triggers ~2 emits, not 3000) so
the frontend's `invalidateQueries` is never spammed.
* Frontend `useUsageEventBridge` listens for the event and invalidates
`usageKeys.all`. Hook is mounted only on `UsageDashboard`, so the
listener is unsubscribed automatically when the user navigates away.
Verification
------------
* `cargo check` passes (existing 25 dead-code warnings in
`commands/misc.rs` are pre-existing and unrelated).
* `tsc --noEmit` passes.
* Manually verified end-to-end: a Codex sync run that imported 3145
entries produced 2 debounced emits, both logged as `emit
usage-log-recorded 成功`, and the dashboard updated within ~200ms.
Behaviour notes
---------------
* `INSERT OR IGNORE` paths (Claude/Codex session sync) only notify when
the row is actually inserted, so dedup-skipped writes don't trigger
empty refreshes.
* Gemini's `INSERT … ON CONFLICT … DO UPDATE` path reuses the existing
`conn.changes() > 0` check and only notifies when token counts truly
changed.
* `rollup_and_prune` notifies once per pruning cycle (at most once per
app start) so the dashboard reflects the new aggregate state.
Co-authored-by: in30mn1a <in30mn1a@users.noreply.github.com>
* feat: add MiMo reasoning_content support for Claude Code proxy
MiMo requires reasoning_content to be passed back in multi-turn
conversations with tool calls. Add "mimo" and "xiaomimimo" to the
preserve_reasoning_content whitelist so thinking blocks are converted
to reasoning_content when proxying Claude Code → MiMo.
Also handle redacted_thinking blocks (encrypted by Claude Code across
turns) by injecting a placeholder to prevent MiMo 400 errors.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: support MiMo reasoning histories
- Fix the PR CI clippy failure for redacted thinking handling.
- Preserve MiMo reasoning content on the OpenAI Chat proxy path.
- Normalize Claude Desktop Anthropic thinking history for MiMo local routes.
- Reject unsupported Claude Desktop direct model remaps.
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Jason <farion1231@gmail.com>
Codex provider switches now only write config.toml for third-party providers,
injecting the API key as experimental_bearer_token. The user's auth.json
(ChatGPT OAuth tokens) is preserved. Official providers with login material
still write auth.json normally. Backfill restores bearer tokens into stored
provider auth.OPENAI_API_KEY to maintain canonical shape.
Harden the proxy module against panics by removing optimistic
unwrap()/expect() calls in favor of pattern matching and graceful
fallbacks:
- copilot_optimizer/cache_injector: bind Value::Array/String directly
instead of is_array()+as_array().unwrap(); use is_none_or and in-place
string mutation
- hyper_client: gate the raw-write path with if-let + filter instead of
has_cases + unwrap()
- gemini_shadow: recover poisoned RwLock via into_inner() rather than
panicking, with warn logging
- streaming_codex_chat: replace expect("tool state exists") with
let-else (return/continue)
- merge_tool_results: early-return when messages is absent
- sse: fall back to from_utf8_lossy on the UTF-8 boundary slice
No behavior change on the happy path; all proxy tests pass.
No-argument tool calls produced empty argument strings that passed
through both the streaming response path and the request transform
verbatim. Strict OpenAI-compatible upstreams (e.g. Minimax) reject
`arguments: ""` with a 400 "invalid function arguments json string",
while lenient ones (OpenAI, Kimi) silently treat it as an empty object.
Add canonicalize_tool_arguments / canonicalize_tool_arguments_str helpers
that coerce empty/whitespace arguments to "{}" and apply them at the four
tool-call argument sites (streaming finalize + three request/response
transform paths). Leave function_call_output content untouched, where an
empty string is valid.
Auto-detect each Chat-routed Codex provider's reasoning interface from
its name, base URL, and model, then inject the matching thinking
parameter without manual configuration:
- Platform-first inference (OpenRouter, SiliconFlow) overrides model
rules, since the same model exposes different reasoning controls
depending on the hosting platform.
- Effort tiers are forwarded only to providers that support them
(DeepSeek, OpenRouter, and StepFun's step-3.5-flash-2603); on/off-only
providers (Kimi, GLM, Qwen, MiniMax, MiMo, SiliconFlow) drop the level
instead of sending a field the upstream rejects.
- OpenRouter uses the native reasoning:{effort} object, clamps max to
xhigh (its enum has no max), and forwards an explicit effort:"none" so
reasoning can be turned off.
- StepFun falls back to inference so per-model effort support is honored
(the static preset would have forced effort on step-3.5-flash too).
Includes the Codex provider-form reasoning controls, i18n strings
(zh/en/ja), and response-side reasoning extraction.
Responses-to-Chat conversion did not set stream_options.include_usage,
so OpenAI-compatible upstreams (kimi/MiniMax, etc.) omitted the trailing
usage chunk in streaming mode. This caused token/cost/cache stats to be
recorded as zero for third-party streaming requests routed through the
Codex Chat path.
Inject include_usage when the request is streaming, merging into any
client-provided stream_options instead of overwriting it. Add tests for
the streaming, non-streaming, and merge cases.
Thinking models like kimi/Moonshot and DeepSeek reject any assistant
message carrying tool_calls without a non-empty reasoning_content. When
cross-turn history recovery misses (proxy restart drops the in-memory
cache, ambiguous call_id, or a turn produced no reasoning upstream), the
converted assistant message has none and the whole request fails with
'reasoning_content is missing in assistant tool call message'.
Add a final-stage backfill that runs after all input items are
processed, so genuine trailing reasoning items still attach first and a
placeholder is only injected when none remains. Mirrors the placeholder
behavior in transform::anthropic_to_openai_with_reasoning_content.
The Codex Chat-to-Responses bridge already rewrites successful upstream
responses to the Responses shape, but the error branch in
handle_codex_chat_to_responses_transform passed Chat-shaped error bodies
through untouched (MiniMax base_resp, raw OpenAI Chat error, text/plain
"Unauthorized" pages, etc.), leaving Codex clients unable to recognize the
error.
Add chat_error_to_response_error in transform_codex_chat to regularize all
upstream error shapes into the standard {error: {message, type, code, param}}
envelope, then wire it through a new handle_codex_chat_error_response that
preserves the original HTTP status code. Non-JSON error bodies (HTML, plain
text) are wrapped as Value::String and truncated to 1KB at a UTF-8 char
boundary to keep diagnostic context without flooding the response.
Also fix a pre-existing append-vs-insert pitfall in three rebuilt-body
branches (Claude transform, Codex Chat normal, Codex Chat error): http
Builder::header is append, so leaking the upstream Content-Type produced
two Content-Type headers when the rewritten body was JSON. Remove the
upstream value before writing application/json.
Codex Chat providers (apiFormat=openai_chat, e.g. DeepSeek, MiniMax, Kimi)
were incorrectly probed via /v1/responses by Stream Check, which the upstream
rejects. Detect chat routing via codex_provider_uses_chat_completions and
issue the probe against /chat/completions with a Chat-shaped body.
Also align URL fallback order with the production CodexAdapter::build_url:
origin-only base URLs (https://api.deepseek.com) must hit /v1/<endpoint>
first; bare /<endpoint> only as a fallback. The previous order made any
non-404 error on the bare path (401/400/405) flag the provider as down even
though /v1/<endpoint> would have succeeded.
- Lift origin-only detection into proxy::providers::is_origin_only_url so
both build_url and Stream Check share a single source of truth.
- Collapse resolve_codex_stream_urls and resolve_codex_chat_stream_urls into
a generic resolve_codex_endpoint_urls(base_url, is_full_url, endpoint),
which also gives the Responses path a symmetric "full endpoint without
is_full_url flag" fallback for free.
- Restrict reasoning_effort propagation in the Chat branch to OpenAI o-series
models, mirroring transform_codex_chat's runtime check.
MiniMax's OpenAI-compatible chat endpoint strict-rejects any non-leading
role=system message with "invalid params, chat content has invalid message
role: system (2013)". The Codex client uses role=developer (and occasionally
role=system) to inject collaboration_mode / permissions / skills blocks
mid-conversation, and responses_role_to_chat_role maps both to chat's
system role. The converted messages array therefore frequently contained
system entries past index 0, which DeepSeek and OpenAI tolerate but MiniMax
flags as 2013.
Collapse all system messages into a single leading system message before
returning the chat request. The rewrite preserves every system fragment
(joined by "\n\n" in original order) and leaves non-system messages
untouched, so it is lossless for permissive backends as well.
- Add collapse_system_messages_to_head in transform_codex_chat.rs.
- Run it on the messages vector at the end of responses_to_chat_completions
before serializing.
- Cover the new path with two unit tests: one repros the MiniMax-shaped
input (developer items between users) and asserts no system role past
index 0; the other verifies non-system order is preserved and content
is joined with "\n\n".
When Codex client sends a model from the catalog (e.g., user selected via /model),
preserve that choice instead of always replacing with config.toml's default model.
- Check if request model exists in modelCatalog before falling back to config model
- Remove CODEX_CHAT_CLIENT_MODEL constant (no longer needed)
- Add test coverage for catalog model preservation
- Add Codex provider API format selection and model mapping for Chat-only upstreams.
- Convert Codex Responses requests to Chat Completions and rebuild Chat responses as Responses output.
- Preserve reasoning_content across non-streaming, streaming, tool calls, and previous_response_id follow-ups.
- Add a bounded Codex Chat history cache for restoring tool calls before tool outputs.
- Cover Chat bridge, compact routing, streaming, and history recovery with focused tests.
- Split leading <think> blocks from Chat content into Responses reasoning output.
- Keep assistant text output free of inline thinking tags.
- Support streamed inline thinking blocks before normal text deltas.
- Add regression coverage for MiniMax-style inline thinking responses.
- Add a Codex API format selector and routing badge for Chat Completions providers.
- Convert Codex Responses requests to upstream Chat Completions when routing is required.
- Convert Chat Completions JSON and SSE responses back to Responses format.
- Keep generated Codex wire_api values on Responses for Codex compatibility.
- Add i18n labels, provider metadata handling, and focused conversion tests.
- Provider: add uses_managed_account_auth / is_github_copilot helpers
to identify managed-account providers (GitHub Copilot / Codex OAuth)
- ProxyService: choose auth policy by provider type when taking over
Claude Live config. Managed accounts drop token env keys and write
only the ANTHROPIC_API_KEY placeholder; other providers keep the
existing ANTHROPIC_AUTH_TOKEN fallback behavior
- Forwarder: add outbound guard that refuses to send the PROXY_MANAGED
placeholder upstream to *.githubcopilot.com and chatgpt.com
/backend-api/codex
- Add unit tests covering detection, injection, and the outbound guard
* fix(gemini-native): resolve functionResponse.name and thought_signature replay for synthesized tool call IDs
Two related bugs in the Gemini Native format conversion layer:
1. **functionResponse.name resolution** (422 error): When Gemini's parallel
function calls omit the id field, cc-switch synthesizes gemini_synth_*
IDs. These are stored in the shadow store but can be lost in long sessions,
causing subsequent tool_result blocks to fail. Fix: pre-scan all assistant
messages in the request body to seed the tool_name_by_id map, and add a
last-resort fallback that scans the current content array for matching
tool_use blocks.
2. **thought_signature replay** (400 error): The Anthropic Messages format
strips thoughtSignature from tool_use blocks, but Gemini requires it on
every functionCall in multi-turn tool-use exchanges. Fix: build a
thought_signature_by_id map from shadow turns and attach thoughtSignature
when converting tool_use back to functionCall.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* style: run cargo fmt on transform_gemini.rs
---------
Co-authored-by: Tiancrimson <tiancrimson@gmail.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Jason <farion1231@gmail.com>
- Mark `should_force_identity_encoding` as test-only.
- Keep runtime forwarding behavior unchanged.
- Verified with local CI checks and no-bundle Tauri build.
- Forwarder buffers non-streaming bodies and primes streaming first
chunk before signaling success, so body timeouts and SSE first-chunk
failures route through the circuit breaker instead of being recorded
as success on response-header arrival
- Atomic enable-failover: switch to P1 before persisting the flag, and
roll back auto-added queue entries when the switch is rejected
(e.g. official providers)
- Hot-reload circuit breaker config on per-app proxy config change
instead of waiting for a proxy restart
- FailoverToggle / FailoverQueueManager / AutoFailoverConfigPanel
require proxy takeover for the active app; the backend command also
rejects enabling when takeover is off
- ProviderHealthBadge consumes the backend is_healthy flag instead of
hardcoding the 5-failure threshold
Cleanup:
- impl From<&AppProxyConfig> for CircuitBreakerConfig and use it from
the command layer
- Collapse three identical TabsContent blocks into a single map
* model pricing routing: extend prefix-match families (gpt-/o1-o5/
gemini-/deepseek-/qwen-/glm-/kimi-/minimax-) with per-family dash
thresholds so short base IDs like gpt-5 no longer mis-match
gpt-5-mini; strip ISO and 8-digit date suffixes via UTF-8-safe
byte matching so claude-haiku-4-5-20251001 falls back to
claude-haiku-4-5 pricing
* SSE collector: SseUsageFinishGuard (RAII) guarantees finish() on
early return or panic; AtomicBool fast path lets push() skip the
Mutex once first-event time is recorded
* validation: shared validate_cost_multiplier / validate_pricing_source
helpers across DAO and service layers; PRICING_SOURCE_RESPONSE /
PRICING_SOURCE_REQUEST constants replace string literals; price
fields in update_model_pricing now reject empty / non-decimal /
negative input before INSERT
* backfill: add backfill_missing_usage_costs_for_model so a single
price edit only scans matching rows instead of the full log table;
startup backfill remains full-scan
* session_usage{,_codex,_gemini}: share find_model_pricing helper from
usage_stats; metadata_modified_nanos centralizes mtime precision
* frontend: NON_NEGATIVE_DECIMAL_REGEX + isNonNegativeDecimalString
replace three copies of the same multiplier regex; isUnpricedUsage
surfaces zero-cost rows that have usage tokens (cached per row to
avoid double evaluation); invalidate usageKeys.all on pricing mutate
so backfilled rows refresh
* stream_check: thread Result from get_auth_headers via map_err so
the workspace builds again
* forwarder: scope rectifier / budget-rectifier flags per-provider so
failover can still apply rectification on the next attempt
* forwarder: categorize before record_result; route NonRetryable and
ClientAbort through release_permit_neutral so client-side failures
don't pollute circuit breaker or DB health
* handler_context: parse Gemini model from uri.path() and strip both
?query and :action verb defensively in extract_gemini_model_from_path
* forwarder + response_processor + handlers: introduce
ActiveConnectionGuard (RAII) so active_connections decrement covers
the full streaming body lifetime, not just response headers
* claude_desktop_config: use sort_by_key to clear the clippy gate
The signature (RECT-003) and budget (RECT-012) rectifier branches each
carried ~50 lines of identical "provider error -> record + continue /
client error -> release permit + return" handling. The only piece that
varied between them was a log label ("整流" vs "budget 整流").
Move the shared logic into RequestForwarder::handle_rectifier_retry_failure
that returns Option<ForwardError> — None means "continue to the next
provider", Some(err) means "terminal failure, return to the client".
Each call site shrinks from ~50 lines to ~17, drops one level of
indentation, and the two branches now provably cannot drift apart.
forwarder.rs nets ~40 lines smaller.
claude.rs and gemini.rs each defined an identical `hv` closure that wrapped
`HeaderValue::from_str` into a ProxyError::AuthError result, and codex.rs
spelled the same conversion out inline. /simplify reviewers flagged this
as drift-prone copy-paste.
Move the conversion into a single `pub fn auth_header_value` in
providers/adapter.rs and have the three adapters import it locally. Same
error wording everywhere, one place to update if HeaderValue semantics
ever change.
The forwarder used to call client.post(&url) / http::Method::POST in
both the reqwest and hyper paths, and the Gemini route table only
registered POST /v1beta/*. As a result anything the Gemini SDK / CLI
sent as GET (models list, models/<id> info) hit a 404 at the router
and bypassed the local proxy's stats, rectifiers, and failover.
Thread the request method end-to-end:
- ProviderAdapter forwarder API now takes the http::Method by reference
per attempt and dispatches client.request(method, &url) for reqwest
and method.clone() for the hyper raw path.
- All five callers in handlers.rs (handle_messages_for_app for Claude /
Claude Desktop, handle_chat_completions, handle_responses,
handle_responses_compact, handle_gemini) pull the method out of the
incoming axum::extract::Request and pass it on.
- handle_gemini tolerates an empty body (GET endpoints have none) and
the forwarder skips serializing / sending a body for GET / HEAD —
attaching JSON to a GET makes Gemini reject the request.
- server.rs swaps the Gemini routes to any(handle_gemini) so the same
handler handles GET / POST / PUT / DELETE, and adds /gemini/v1/*
for the GA path version.
Three statistics-shape issues fixed together so the dashboard reflects
client requests, not provider attempts:
1. active_connections never moved off zero — the field had no caller in
the entire crate. Wrap forward_with_retry into a thin entry point
that saturating_add(1) on enter and saturating_sub(1) on exit; every
inner return path is covered automatically.
2. total_requests counted attempts, not requests. A single client call
that failed over P1 -> P2 -> success was recorded as
total=2 / success=1 -> 50% success rate. Move the increment and the
last_request_at refresh into the wrapper so they fire once per
client request regardless of how many providers were tried.
3. current_provider / current_provider_id stay inside the inner loop
because they are intentionally per-attempt ("what am I trying right
now?") — moving them would break the live-failover indicator.
Refactor: split forward_with_retry into a public wrapper + private
forward_with_retry_inner. Every existing `return Err(...)` inside inner
remains correct because the wrapper always runs the decrement on its
return.
The UI has exposed "请求失败时的重试次数 (0-10, default 3)" since the
auto-failover panel was added, but the value was silently dropped —
RequestForwarder never received it and the per-provider loop walked the
whole list regardless. From the user's perspective the setting was
inert.
Thread AppProxyConfig.max_retries through create_forwarder into
RequestForwarder, derive max_attempts = max_retries + 1 (so max_retries=0
matches the UI copy "0 retries" = single attempt), and break the loop
once attempts hit the cap. The check is placed before the circuit
breaker allow-permit so an over-cap iteration does not waste a HalfOpen
probe slot.
When auto-failover is disabled we also force max_retries to 0, mirroring
how timeouts already bypass in that mode — "no failover" should mean
"one provider, one try", not "limited retries against the same list".
The Chat-Completions transformer used to forward tool_choice verbatim,
but the two APIs disagree on shape:
Anthropic "any" | {"type":"tool","name":"X"}
OpenAI Chat "required" | {"type":"function","function":{"name":"X"}}
Pass-through made the upstream return 400 for any tool-forcing client
(Claude Code, Copilot, etc.). The Responses-API transformer already had
the equivalent map_tool_choice_to_responses helper; this commit adds a
sibling map_tool_choice_to_chat with the chat-specific *nested* function
selector and five regression tests covering string / object × any /
auto / none / tool.
The two helpers are intentionally not merged: the difference between
flat and nested function selectors is exactly what the original bug
was, so keeping them as separate self-documenting functions reduces the
chance of the same regression returning.
Two related changes to make per-provider failover behave correctly.
1. Bucket UpstreamError by status code in categorize_proxy_error.
The old "every UpstreamError is Retryable" rule meant a malformed
client request (400 / 422) would be replayed against every provider
in the queue: errors amplified N-fold, the circuit breaker accrued
unwarranted failure counts, and quota was burned. Now
400 / 405 / 406 / 413 / 414 / 415 / 422 / 501 are NonRetryable since
the request itself is wrong and no provider will accept it.
401 / 403 / 404 / 408 / 409 / 429 / 451 and all 5xx remain Retryable
because the next provider may carry a different key, quota, region,
or model mapping.
2. Make the rectifier-retry path participate in failover.
Both the signature (RECT-003) and budget (RECT-012) rectifier branches
used to "return Err(...)" after the retry failed, short-circuiting the
per-provider loop. A provider-side failure (5xx / Timeout /
ForwardFailed) now records the circuit breaker, accumulates into
last_error / last_provider, and "continue"s to the next provider —
matching the normal Retryable arm. Client-side failures still return
immediately since a different provider cannot fix a malformed payload.
split('/') strips the slashes, so find(|s| s.starts_with("models/")) never
matched any segment and request_model fell through to "unknown" for every
Gemini call, poisoning usage records, per-request billing, and logs.
Match the literal "models" segment and take the next one, stripping any
:action suffix and query string. The extraction is now a pub(crate) free
function so it can be unit-tested directly; seven regression tests cover
action suffixes, dotted versions, the /gemini/ proxy prefix, query
strings, the bare list endpoint, and missing-segment paths.
User-pasted API keys can contain control chars or CR/LF that make
HeaderValue::from_str return Err; the previous unwrap inside every
adapter turned such input into a process-wide panic instead of a request
error. The trait now returns Result<_, ProxyError>; Claude/Codex/Gemini
impls propagate ProxyError::AuthError so the client sees a 401 with the
underlying parse error instead of a crash. Adds a regression test that
pastes a CRLF-containing key and asserts AuthError.
- Split CostCalculator into per-app cache semantics: Anthropic's
input_tokens is already fresh input, while Codex/Gemini include
cached tokens in their prompt count. The old shared formula
double-subtracted cache_read for Claude, under-billing input cost.
- Backfill now reads cost_multiplier from the per-log snapshot column
instead of re-querying providers.meta, so historical rows are no
longer rewritten with the current multiplier.
- Move the "pricing not found" warn out of find_model_pricing_row;
emit it only when a brand new log is written, and skip placeholder
models (unknown / empty / null / none) entirely.
- Broaden model id normalization: strip namespace prefixes
(anthropic./openai./global./bedrock.), bedrock-style -vN suffixes,
reasoning effort suffixes (-low/-medium/-high/-xhigh/-minimal),
Claude Desktop's claude-<non-anthropic> wrapper, dot-to-dash for
Claude, and try a LIKE prefix match for Claude short route ids
(e.g. claude-haiku-4-5 -> claude-haiku-4-5-20251001).
- Fall back to request_model when the stored model is missing, so
early Codex session rows with model=unknown can still be priced.
- Replace the four flat env inputs with a Sonnet/Opus/Haiku role table.
Each row exposes ANTHROPIC_DEFAULT_*_MODEL plus a new display name
field ANTHROPIC_DEFAULT_*_MODEL_NAME, and Sonnet/Opus gain a
"Declare 1M" checkbox that toggles the [1M] suffix.
- Strip the [1M] context-capability marker before forwarding non-Copilot
requests upstream. Copilot keeps its existing [1m]->-1m normalization.
- Claude Desktop import now consumes ANTHROPIC_DEFAULT_*_MODEL_NAME as
label_override, closing the Claude Code -> Claude Desktop displayName
pipeline; add_route's merge logic is shared between hashmap branches.
- Unify the [1M] marker as ONE_M_CONTEXT_MARKER across
claude_desktop_config and proxy::model_mapper; rename the strip
helper to strip_one_m_suffix_for_upstream.
- Collapse useModelState's seven duplicated useState initializers and
the useEffect parse block into a single parseModelsFromConfig call.
- Add tests/hooks/useModelState.test.tsx and a Claude Desktop import
test covering Kimi K2 -> label_override. i18n (en/ja/zh) updated.
- Guard debug body serialization with `log::log_enabled!`; previously
serialized the filtered body to a throwaway String on every forward,
even with debug logging off.
- Skip SSE parse + UTF-8 buffer loop when no usage collector and debug
is off; the per-chunk `serde_json::from_str::<Value>` ran even in
pure passthrough mode.
- Add cheap per-app SSE event pre-filter (string `contains`) so usage
collectors only parse events that could contain usage (e.g. Claude
`message_start` / `message_delta`).
- Skip non-streaming response body JSON parse when usage logging is
disabled.
- Move `ProviderRouter::record_result` off the success response path
via `tokio::spawn` for non-HalfOpen state; that call internally does
`get_proxy_config_for_app` + `update_provider_health`, two SQLite
ops that previously blocked TTFB.
Also: dedupe `usage_logging_enabled` (was duplicated in handlers.rs)
and merge `SseUsageCollector::{new, new_filtered}` into a single
constructor that takes `Option<StreamUsageEventFilter>`.
prompt_cache_key was falling back to provider.id when the client did not
supply a session, which collapsed every conversation onto a single key
and defeated upstream prefix caching. Only emit the key when a real
client-provided session/thread identity is available; otherwise let the
upstream use its default matching behaviour.
Additional fixes that affect cache stability:
- Canonicalise (sort) JSON keys in outgoing request bodies and in
tool_call arguments / tool_result content so semantically identical
requests produce identical byte sequences for upstream prefix caches.
- Exempt JSON Schema property maps (properties, patternProperties,
definitions, \$defs) from the underscore-prefix filter so user-defined
schema keys like _id and _meta survive.
- Add a [CacheTrace] debug log with stable hashes for instructions,
tools, input and include to help diagnose cache misses.
- Thread session_id into the usage logger for request correlation.
- Apply rustfmt diffs in claude_desktop_config.rs
- Allow needless_return on current_platform_paths (cfg-mirrored arms)
- Allow too_many_arguments on RequestForwarder::forward
- Replace `let mut + reassign` with struct literals in tests
(settings, backup, provider, response_processor)
- Use Path::new instead of PathBuf::from to fix cmp_owned in misc tests
- Replace 3.14 with 3.5 in config test to avoid approx_constant lint
Adds a new ClaudeDesktop AppType that writes Claude Desktop's third-party
inference profile under configLibrary/, sharing _meta.json with other
launchers (Ollama-compatible) so cc-switch can coexist with them.
Two switch modes:
- direct: provider already exposes claude-* / anthropic/claude-* model
ids on Anthropic Messages, Claude Desktop connects to it directly.
- proxy: cc-switch's local proxy acts as the inference gateway,
presenting only claude-* route names to Claude Desktop and mapping
them to real upstream models. Required after Anthropic restricted
Claude Desktop to claude-family ids.
Backend:
- New module claude_desktop_config with snapshot/rollback, official seed
bypass, /claude-desktop/v1/{models,messages} routes, and a single
source of truth for default proxy routes.
- Gateway token persisted in SQLite, validated on every proxied request.
- get_claude_desktop_status surfaces drift signals (stale models,
missing routes, proxy stopped, base URL mismatch, missing token).
Frontend:
- Slim ClaudeDesktopProviderForm independent from ProviderForm,
controlled by a top-level appId guard.
- ProviderList banner consumes the status query (5s polling) and
renders actionable diagnostics.
- ClaudeDesktopRouteToggle in the header to start/stop the local
gateway without touching takeover state.
- Three-locale i18n synchronised.
The hyper raw-write path preserves original header casing but rebuilds
TCP+TLS on every request — there is no connection pool — which was the
root cause of slow reverse-proxy throughput.
Only Anthropic-native requests actually need exact header-case
preservation. Route OpenAI/Copilot/Codex/Gemini/codex_oauth requests
through the pooled reqwest client (pool_max_idle_per_host=10,
tcp_keepalive=60s) instead, so warm connections get reused.
Streaming requests get a precise first-byte timeout via
tokio::time::timeout around reqwest's send() (which resolves on
response headers), with the body phase handed off to response_processor.
The streaming-detection helper now also covers Gemini SSE endpoints
and Accept: text/event-stream, not just body.stream.
Anthropic SDK assigns distinct semantics to the two env vars:
- ANTHROPIC_API_KEY -> x-api-key
- ANTHROPIC_AUTH_TOKEN -> Authorization: Bearer
The Claude adapter previously collapsed both into AuthStrategy::Anthropic
and then emitted Authorization: Bearer regardless, breaking strict
Anthropic-protocol endpoints (Anthropic official, Cloudflare AI Gateway,
OpenCode Go, DashScope) and silently overriding the user's intended auth
scheme.
- claude::extract_auth: infer strategy from env var name
(ANTHROPIC_AUTH_TOKEN -> ClaudeAuth, ANTHROPIC_API_KEY -> Anthropic),
matching the precedence already used by extract_key.
- claude::get_auth_headers: split the Anthropic arm so it emits
x-api-key, while ClaudeAuth and Bearer continue to use Bearer.
- stream_check: reuse ClaudeAdapter::get_auth_headers as the single
source of truth, replacing the prior "always Bearer + maybe x-api-key"
double injection that produced auth conflicts and false-negative
health checks.
- Cover each strategy -> header mapping and env-var precedence with
new unit tests in claude.rs.
Refs #2368, #2380
Claude Code injects a dynamic `x-anthropic-billing-header` line at the
start of `system` content. Its rotating `cch=` token was forwarded into
OpenAI Responses `instructions` and Chat system messages, which broke
upstream prefix prompt cache reuse — a stable ~95k-token prefix was
getting re-charged on every request.
Strip only the leading occurrence in both anthropic_to_openai and
anthropic_to_responses; later occurrences are preserved so user-authored
prompt text containing the same string is not lost.
Haoyu Wang