fix: Agnes/LongCat OpenAI-only capabilities - #92
Merged
Conversation
…trate - Add Strict field to responsesTool struct, set strict: true on all tools - Add normalizeToolParams to inject additionalProperties: false for object schemas - Restructure responsesTool to flat format (name/description/parameters at top level) - Add Store field to responsesRequest - Expand response types with Status, Error, IncompleteDetails - Add bufio import for SSE reader Per Concentrate API docs: strict mode requires additionalProperties: false.
Previously required CONCENTRATE_FETCH_PRICING=true env var, which meant most users saw "free" for all models. Now pricing is fetched by default and cached for 24 hours. Falls back gracefully to no pricing if API key is unavailable.
Map all Concentrate API fields from ChatOptions to responsesRequest: - include, routing, max_tool_calls, context_management - safety_identifier, background, truncation, conversation Per: https://concentrate.ai/docs/api-reference/endpoint/request-parameters
Revert manual go.mod edits; workspace handles replacement.
- Register Agnes in provider registry (ChatPreference: 25, free tier) - Add FetchAgnes live fetcher calling /v1/models - Add AGNES_API_KEY credential mapping - Add Agnes fields to ProviderConfig - Update tests for 24 providers (was 23) - Agnes is free with RPM limits, no per-token pricing Base URL: https://apihub.agnes-ai.com/v1
Previously only Concentrate got special treatment for unknown function-calling capability. Now all providers with unknown capabilities (like Agnes AI free tier) are assumed to support tools. Better to try and fail than block models from being used. Only explicit unsupported blocks a model. Updates test to reflect new behavior: unknown = assumed supported.
- Add agnes-direct case in providerForDeployment switch - Add Agnes to OpenAICompatibleRuntimeProfiles and RuntimeProviderProfiles - Add ProviderAgnes constant - Reuses OpenAI adapter (OpenAICompat) for Agnes OpenAI-compatible API - Base URL: https://apihub.agnes-ai.com/v1 Fixes no route configured error when selecting Agnes models.
- Add OmitMaxTokens compat flag to skip max_tokens for pre-deduction providers - Add AgnesCompat config with OmitMaxTokens=true to match curl behavior - Register AgnesCompat in init() for OpenAI-compatible providers - Guard max_tokens in buildRequestBase by OmitMaxTokens check - Switch agnes-direct deployment from OpenAICompat to AgnesCompat - Mark Agnes pricing as unknown (PricingUnknown) instead of free (pre-deduction) - Fix tests for new pricing sentinel behavior
- Add agnesProviderStages with anthropic-direct fallback when both are configured - Add agnes-direct to deploymentOwnerProviderID - Add insufficient_user_quota/pre-deduction patterns to ShouldTryNextDeployment so Agnes balance errors trigger fallback to Anthropic
LongCat AI provides LongCat-2.0 model with 1M token context. Supports both Token Pack and Pay-As-You-Go plans via same endpoint. Endpoint: https://api.longcat.chat/openai/v1 Uses OpenAICompat config. Added live model fetcher, routing, and profiles.
LongCat supports both OpenAI and Anthropic API formats. If the OpenAI-compatible endpoint fails, fall back to the Anthropic endpoint at https://api.longcat.chat/anthropic/v1/messages
LongCat should only fall back between its own OpenAI and Anthropic endpoints, not to OpenRouter.
Reorder all provider SortOrder values to alphabetical by DisplayName. Fixes duplicate SortOrder values that were introduced during renumbering.
Drop cross-provider and longcat-anthropic fallbacks, map longcat-direct credentials, and enrich official context/tools/vision/thinking metadata from provider docs when /models omits it. Co-authored-by: Cursor <cursoragent@cursor.com>
… and update credential resolve test
Patel230
added a commit
that referenced
this pull request
Jul 30, 2026
* feat: strict mode compliance and tool schema normalization for Concentrate - Add Strict field to responsesTool struct, set strict: true on all tools - Add normalizeToolParams to inject additionalProperties: false for object schemas - Restructure responsesTool to flat format (name/description/parameters at top level) - Add Store field to responsesRequest - Expand response types with Status, Error, IncompleteDetails - Add bufio import for SSE reader Per Concentrate API docs: strict mode requires additionalProperties: false. * fix: fetch Concentrate pricing by default instead of opt-in Previously required CONCENTRATE_FETCH_PRICING=true env var, which meant most users saw "free" for all models. Now pricing is fetched by default and cached for 24 hours. Falls back gracefully to no pricing if API key is unavailable. * feat: implement Concentrate Responses API extended parameters Map all Concentrate API fields from ChatOptions to responsesRequest: - include, routing, max_tool_calls, context_management - safety_identifier, background, truncation, conversation Per: https://concentrate.ai/docs/api-reference/endpoint/request-parameters * fix: restore go.mod after workspace testing Revert manual go.mod edits; workspace handles replacement. * feat: add Agnes AI as free OpenAI-compatible provider - Register Agnes in provider registry (ChatPreference: 25, free tier) - Add FetchAgnes live fetcher calling /v1/models - Add AGNES_API_KEY credential mapping - Add Agnes fields to ProviderConfig - Update tests for 24 providers (was 23) - Agnes is free with RPM limits, no per-token pricing Base URL: https://apihub.agnes-ai.com/v1 * fix: treat unknown capabilities as supported for all providers Previously only Concentrate got special treatment for unknown function-calling capability. Now all providers with unknown capabilities (like Agnes AI free tier) are assumed to support tools. Better to try and fail than block models from being used. Only explicit unsupported blocks a model. Updates test to reflect new behavior: unknown = assumed supported. * fix: add Agnes deployment routing and runtime profile - Add agnes-direct case in providerForDeployment switch - Add Agnes to OpenAICompatibleRuntimeProfiles and RuntimeProviderProfiles - Add ProviderAgnes constant - Reuses OpenAI adapter (OpenAICompat) for Agnes OpenAI-compatible API - Base URL: https://apihub.agnes-ai.com/v1 Fixes no route configured error when selecting Agnes models. * feat: add Agnes AI provider with OmitMaxTokens compat fix - Add OmitMaxTokens compat flag to skip max_tokens for pre-deduction providers - Add AgnesCompat config with OmitMaxTokens=true to match curl behavior - Register AgnesCompat in init() for OpenAI-compatible providers - Guard max_tokens in buildRequestBase by OmitMaxTokens check - Switch agnes-direct deployment from OpenAICompat to AgnesCompat - Mark Agnes pricing as unknown (PricingUnknown) instead of free (pre-deduction) - Fix tests for new pricing sentinel behavior * feat: add Agnes-to-Anthropic fallback routing - Add agnesProviderStages with anthropic-direct fallback when both are configured - Add agnes-direct to deploymentOwnerProviderID - Add insufficient_user_quota/pre-deduction patterns to ShouldTryNextDeployment so Agnes balance errors trigger fallback to Anthropic * feat: add LongCat AI provider (OpenAI-compatible endpoint) LongCat AI provides LongCat-2.0 model with 1M token context. Supports both Token Pack and Pay-As-You-Go plans via same endpoint. Endpoint: https://api.longcat.chat/openai/v1 Uses OpenAICompat config. Added live model fetcher, routing, and profiles. * feat: add longcat-anthropic fallback deployment LongCat supports both OpenAI and Anthropic API formats. If the OpenAI-compatible endpoint fails, fall back to the Anthropic endpoint at https://api.longcat.chat/anthropic/v1/messages * fix: remove OpenRouter fallback from LongCat routing LongCat should only fall back between its own OpenAI and Anthropic endpoints, not to OpenRouter. * fix: alphabetical SortOrder values for all providers Reorder all provider SortOrder values to alphabetical by DisplayName. Fixes duplicate SortOrder values that were introduced during renumbering. * fix: keep Agnes and LongCat OpenAI-only with full live capabilities Drop cross-provider and longcat-anthropic fallbacks, map longcat-direct credentials, and enrich official context/tools/vision/thinking metadata from provider docs when /models omits it. * fix: complete changes for Agnes/LongCat OpenAI-only capabilities * fix: remove invalid ChatOptions field accesses in concentrate adapter and update credential resolve test * fix: resolve golangci-lint issues for gocritic nilValReturn and unused transportError * style: format test_helpers_test.go with gofumpt ---------
Patel230
added a commit
that referenced
this pull request
Jul 31, 2026
Patel230
added a commit
that referenced
this pull request
Jul 31, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Complete changes for Agnes and LongCat AI provider OpenAI-only routing