- Preserve pending tool use IDs across message-level system reminders to maintain tool call and result alignment.
- Merge adjacent Gemini request contents for consecutive user and system reminder turns.
- Relax tool response reordering condition and preserve non-response parts in Antigravity executor.
- Align tool results and buffer pending system reminders in OpenAI translator.
Closes: #5354
Track when executor calls cross an upstream transport boundary and use that
signal to keep model/provider errors from being replaced by later local
preparation, selection, or internal failures.
Mark HTTP, websocket, relay, and usage-tracked transports as upstream
attempts, while avoiding marks for local validation, logging, missing
sessions, and successful websocket handshakes before request send.
Parse relative auth expiry metadata and adjust Antigravity refresh timing.
- Add `AlignClaudeToolResults` to order `tool_result` blocks to match the preceding `tool_use` IDs while preserving other content parts.
- Apply tool result alignment in Claude-to-Gemini and Claude-to-Antigravity request translators.
- Preserve mixed non-response parts when normalizing and reordering parallel function responses in Antigravity executor.
Closes: #5199
* fix(executor): prepend empty user turn for model-first requests targeting Gemini/Antigravity (#4959)
When forwarding sliced conversation histories or tool calls across OpenAI Responses,
OpenAI Chat Completions, Claude Messages, and native Gemini, native Gemini and Antigravity
Gemini endpoints require that conversation contents begin with a user turn.
Normalize leading turns at the executor boundary rather than the translator layer:
- Prepend an empty user turn ({"role":"user","parts":[{"text":""}]}) for Gemini, Gemini Vertex,
AI Studio, and Antigravity Gemini generation and CountTokens requests if the first turn is 'model'.
- Keep Antigravity Claude requests untouched to avoid adapter 400 errors.
- Ensure normalization runs after payload rules so payload index overrides target the original turns.
- Use no-copy GJSON inspection to keep overhead zero on valid user-first requests.
* fix(executor): inject Antigravity leading user after reasoning replay (#4959)
Replay can insert a model functionCall at contents[0] for sliced
tool-result history. Run the empty-user prepend on the final
requestPayload, after sanitize and prepareAntigravityGeminiReasoningReplayPayload.
Review and live-test follow-ups to the shared upstream transport.
Bound the cache with an LRU that closes idle connections on eviction, so
rotating a credential's proxy or supplying a per-request base transport can
no longer leak pools.
Stop deriving a pool scope from Auth.Label: it is documented as an optional
human readable label for logging and carries no uniqueness guarantee, so two
OAuth identities sharing a label would share one TCP/TLS pool. Prefer a
refresh-token digest, which stays stable across access-token rotation and is
available to refresh requests that run before any access token exists.
Replace a typed-nil *http.Transport taken from the request context. It passes
the interface nil check, so leaving it in place made http.Client fall back to
http.DefaultTransport, which advertises h2 over ALPN and breaks the
HTTP/1.1-only fingerprint.
Only widen pool limits: treat MaxIdleConns == 0 and IdleConnTimeout == 0 as
unlimited, and leave a negative MaxIdleConnsPerHost alone because that is how
an operator disables pooling.
Size the cache for large deployments. An unused entry costs under 1 KB and no
goroutines, whereas evicting a live pool forces a fresh TCP + TLS handshake,
so capacity is not the lever for bounding memory.
gjson.GetBytes copies the matched subtree before returning it, so every read of
request.contents on a multi-megabyte request allocated another copy of it. The
reasoning replay, executor, signature validation and session identity paths all
read whole arrays this way, and hasExplicitSession parsed the entire body with
gjson.ParseBytes on every request.
Use the no-copy helpers for these reads. Each call site only reads from the
result and writes through sjson, which allocates a new buffer, so the payload
stays immutable while the results are alive.
- Removed deprecated interceptor and executor-related methods, including `callRequestInterceptor`, `callResponseInterceptor`, and `callStreamChunkInterceptor`.
- Consolidated unused logic and pruned redundant imports to streamline `adapters.go`.
- No functional changes.
- Updated Kimi and Gemini executors to use `TranslateRequestWithCodexMultiAgentV2` for improved multi-agent v2 optimization.
- Enhanced configuration and API server to support `CodexOptimizeMultiAgentV2`.
- Added codex multi-agent V2 optimizations in `config`, `executor`, and `watcher` modules.
- Introduced `ClaudeInputTokenState` to track and estimate input token usage for Claude message_start events.
- Added `TranslateStreamWithClaudeInputTokens` for token-aware stream translation.
- Updated executors (XAI, Kimi, Gemini) to leverage the new logic.
- Included robust test cases for token counting, JSON validation, and concurrent tokenization scenarios.
- Introduced `SetStringIfDifferent`, `SetBoolIfDifferent`, and similar helpers in `payload_helpers` to avoid unnecessary writes during JSON modifications.
- Simplified message patching in `kimi_executor` for reasoning and tool call adjustments.
- Updated all executors (Gemini, Codex, XAI, Antigravity, etc.) to use the new helpers, enhancing readability and potentially reducing overhead.
- Removed obsolete `filterKimiEmptyAssistantMessages` function in `kimi_executor`.
- Introduced API handlers and executor logic for Google Interactions
- Added request and response transformations for OpenAI and Claude Interactions.
- Integrated Gemini API with Interactions support.
- Updated tests to validate Interactions request parsing and error handling.
- Refactored translator logic for Interactions data flows.
Use the aidev_client parenthetical suffix for default Antigravity UA and
route loadCodeAssist through the short UA. Keep onboardUser on the long UA
with google-api-nodejs-client.
Add executor-scoped replay cache aligned with Codex HOME replay:
Scope, observe SSE/non-stream responses, store normalized thought_signature
and function_call_part items, apply on the next streamGenerateContent
request, and invalidate on invalid signature responses.
Gemini/flash/agent models use HOME replay; native per-part signature
replay is not wired on upstream/dev. Wire non-stream and stream paths
in antigravity_executor and purge expired entries from signature_cache.
Includes unit tests and HOME-provider-replay documentation.
Add executor-scoped replay cache aligned with Codex HOME replay:
Scope, observe SSE/non-stream responses, store normalized thought_signature
and function_call_part items, apply on the next streamGenerateContent
request, and invalidate on invalid signature responses.
Gemini/flash/agent models use HOME replay; native per-part signature
replay is not wired on upstream/dev. Wire non-stream and stream paths
in antigravity_executor and purge expired entries from signature_cache.
Includes unit tests and HOME-provider-replay documentation.
- Deleted `geminicli` provider and related `Apply` logic.
- Removed all translator packages specific to Gemini CLI (Claude, Codex integrations).
- Purged associated test files for Gemini CLI translation.
- Removed `GeminiAuthenticator` and all associated authentication logic (OAuth flows, token handling, refresh logic).
- Deleted internal/executor Gemini OAuth support, including bearer token handling and runtime API logic.
- Purged all tests, configs, and command-line flags specific to Gemini OAuth flows.
- Updated documentation and aliases to reflect Gemini removal.
- Renamed `parseRetryDelay` to `ParseRetryDelay` and `deleteJSONField` to `DeleteJSONField`.
- Updated references in `antigravity_executor` and tests to use the new `helps` package.
- Adjusted import paths and test cases to ensure compatibility with the new location.
- Updated README files to reflect changes in the retry logic references.
- Updated `.github/ISSUE_TEMPLATE/bug_report.md` to remove deprecated Gemini CLI mention.
Use the agy CLI User-Agent family (antigravity/cli/{version} darwin/arm64)
on CPA macOS/arm64 hosts instead of the legacy hub-style antigravity/{version}
string. Resolve the cached version from the CLI auto-updater manifest
(darwin_arm64.json), then the GCS latest pointer, then antigravity-cli GCS
prefix listing, with fallback 1.0.8 when all sources fail.
Update AntigravityUserAgent helpers and executor default UA comment to match.
- Updated Antigravity Credits fallback to handle KV store unavailability as a service error.
- Enhanced signature caching mechanisms with request-time KV access and sliding expiration.
- Added and improved tests for KV client interactions, including error handling and expiration behaviors.
- Introduced `CacheSignatureBestEffort` for non-critical signature caching and clarified function flows with required context.
- Ensured consistent error reporting for missing or unavailable KV stores in various scenarios.
- Replaced direct `homekv` calls with injectable KV client interfaces for `antigravity` and `codex_reasoning_replay` modules.
- Improved error reporting and handling for KV operations, including `KVGet`, `KVSet`, `KVDel`, and `KVExpire`.
- Introduced dedicated fake KV clients for expanded and granular test coverage.
- Added new unit tests to validate KV client behaviors and error scenarios, ensuring robustness and sliding expiration functionality.
Add a native Antigravity WebSearch path for Claude typed WebSearch requests.
Detect Claude Messages requests whose tools are only typed WebSearch tools
(web_search_20250305 / web_search_20260209), and convert them into an
Antigravity requestType=web_search payload instead of sending the request
through the normal tool-calling path.
Preserve the user's requested model. The native path is enabled only when that
Antigravity model is known to support Google Search. Capability data fetched
from Antigravity model info is used only as an enhancement to the local model
registry, not as a replacement for the existing registry fallback behavior.
Unsupported models keep the existing Antigravity request behavior and are not
silently rerouted to another web-search-capable model.
Translate Claude WebSearch request options to the verified Antigravity
googleSearch shape:
- max_uses -> googleSearch.enhancedContent.imageSearch.maxResultCount
- allowed_domains -> googleSearch.includedDomains
Leave blocked_domains and user_location unmapped because the Antigravity
googleSearch request shape has no verified equivalent for them. This avoids
sending speculative fields or pretending unsupported Claude WebSearch options
are enforced upstream.
Translate Antigravity web-search responses back into Claude-compatible output:
server_tool_use blocks, web_search_tool_result blocks, cited text blocks,
grounding URLs, and usage-compatible stream/non-stream responses.
Cover the behavior with tests for request conversion, response conversion,
grounding URL resolution, domain filter mapping, fetched capability hints,
excluded-model handling, and unsupported-model behavior.
- Added an example plugin `host-model-callback` in Go to summarize host model callbacks.
- Implemented `cliproxy_plugin_init`, `cliproxyPluginCall`, and other plugin functions for callback handling.
- Introduced API handlers for `ModelExecution` and `ModelExecutionStream` with support for both streaming and non-streaming requests.
- Included unit tests (`model_execution_test.go`) to validate execution logic and streaming responses.
- Replaced `NewUsageReporter` with `NewExecutorUsageReporter` to include executor type in usage records.
- Updated all executors to use the new reporter implementation.
- Extended `UsageReporter` to track and publish executor type.
- Added tests to validate proper executor type recording and handling.
- Enhanced RedisQueue plugin and payload schema with executor type support.
- Introduced Time-To-First-Token (TTFT) measurement and reporting across major executors.
- Added TTFT calculation to `UsageReporter`, including support for HTTP clients and WebSocket communication.
- Updated tests to validate TTFT tracking in streamed and non-streamed scenarios.
- Ensured integration with `usage` plugin and augmented usage records with TTFT data.
- Introduced `SetTranslatedReasoningEffort` method in `UsageReporter` to capture and log reasoning efforts from translated payloads.
- Updated executors to incorporate the new reporting functionality for handling reasoning efforts across various providers.
- Enhanced logging for thinking level extraction with new helper function `ExtractTranslatedReasoningEffort`.
- Introduced `resolvePayloadRulePaths` function to dynamically resolve rule paths supporting array queries and complex logic.
- Updated payload processing logic (`apply defaults`, `overrides`, `filters`) to handle resolved paths for better flexibility.
- Added helper functions for path parsing, query matching, and logical resolution to improve modularity and reusability.
- Introduced payload condition match logic, including `match`, `not-match`, `exist`, and `not-exist` rules in `PayloadConfig`.
- Enhanced `payloadModelRulesMatch` function to support conditional checks at various levels.
- Added helper methods for evaluating JSON path conditions and values.
- Updated tests to validate new conditional rules against different payload scenarios.
- Added `count` attribute to `homeAuthCount` requests to improve home message batching.
- Enabled usage statistics for home mode by default and added config-level enforcement.
- Adjusted failure logging to include detailed metadata in `UsageReporter`.
- Updated multiple executors to pass error details to `PublishFailure` for better debugging.
- Enhanced unit tests to validate `count` behavior and usage statistics enforcement across components.
- Updated all references from v6 to v7 for `github.com/router-for-me/CLIProxyAPI`.
- Ensured consistency in imports within core libraries, tests, and integration tests.
- Added missing tests for new features in Redis Protocol integration.
All streaming executors use bare channel sends (out <- chunk) inside goroutines
that process upstream SSE responses. When the downstream consumer disconnects
(client timeout, network drop, etc.), these sends block indefinitely, causing
the goroutine and all associated resources (HTTP response body, scanner buffers,
translation state) to leak permanently.
Over time, leaked goroutines accumulate monotonically, leading to RSS growth
from ~30MB to 3.7GB+ and eventual OOM kills on resource-constrained VPS hosts.
Fix: Replace all bare 'out <- ...' sends with:
select {
case out <- ...:
case <-ctx.Done():
return
}
This ensures goroutines terminate promptly when the request context is canceled,
allowing GC to reclaim all associated resources.
Affected executors (9 files, 36+ send sites):
- antigravity_executor.go (5 sites)
- gemini_cli_executor.go (6 sites)
- gemini_vertex_executor.go (6 sites)
- aistudio_executor.go (4 sites)
- gemini_executor.go (3 sites)
- openai_compat_executor.go (3 sites)
- claude_executor.go (4 sites)
- codex_executor.go (2 sites)
- kimi_executor.go (3 sites)
- Introduced `DisableImageGenerationMode` with support for `false`, `true`, and `chat` values.
- Updated payload handling to preserve `image_generation` on images endpoints when `chat` mode is enabled.
- Modified OpenAI image handlers (`ImagesGenerations`, `ImagesEdits`) to respect tri-state logic.
- Added unit tests for `DisableImageGenerationMode` behavior and endpoint-specific handling.
- Enhanced configuration diff logging to support `DisableImageGenerationMode`.
The loadCodeAssist polling call hardcoded the User-Agent to
google-api-nodejs-client/9.15.1. Google Cloud Code returns the
paidTier object WITHOUT the availableCredits array for that UA,
so updateAntigravityCreditsBalance always saw "no credits", set the
hint to Available=false for every Google One AI Ultra account, and
the conductor-level credits fallback could never find a candidate.
Switching to resolveUserAgent(auth) (the same UA used for
streamGenerateContent / generateContent) makes the response include
availableCredits, so the credits hint is populated correctly and the
fallback can actually inject enabledCreditTypes:["GOOGLE_ONE_AI"]
when free tier is exhausted.
Move credits handling from executor-level retry to conductor-level
orchestration. When all free-tier auths are exhausted (429/503), the
conductor discovers auths with available Google One AI credits and
retries with enabledCreditTypes injected via context flag.
Key changes:
- Add AntigravityCreditsHint system for tracking per-auth credits state
- Conductor tries credits fallback after all auths fail (Execute/Stream/Count)
- Executor injects enabledCreditTypes only when conductor sets context flag
- Credits fallback respects provider scope (requires antigravity in providers)
- Add context cancellation check in credits fallback to avoid wasted requests
- Remove executor-level attemptCreditsFallback and preferCredits machinery
- Restructure 429 decision logic (parse details first, keyword fallback)
- Expand shouldAbort to cover INVALID_ARGUMENT/FAILED_PRECONDITION/500+UNKNOWN
- Support human-readable retry delay parsing (e.g. "1h43m56s")
Claude models on antigravity have a 64000 token output limit but
max_tokens from downstream requests was passed through uncapped,
causing 400 INVALID_ARGUMENT from Google when clients sent 128000.
