feat(provider): add Responses API support for custom OpenAI-compatible providers #7304
Conversation
|
The following comment was made by an LLM, it may be inaccurate: No duplicate PRs found |
…e providers
- Register @opencode-ai/openai-compatible as bundled provider with Responses API support
- Add omitMaxOutputTokens option for providers that don't support max_output_tokens parameter
- Default to responses() API, with option to fallback to chat() via useResponsesApi: false
This enables custom providers (like OpenRouter, privnode, etc.) that support
OpenAI's /responses endpoint to work with OpenCode.
Usage in opencode.json:
{
"provider": {
"my-provider": {
"npm": "@opencode-ai/openai-compatible",
"api": "https://my-api.com/v1",
"options": {
"omitMaxOutputTokens": true
}
}
}
}
There was a problem hiding this comment.
Pull request overview
This PR adds Responses API support for custom OpenAI-compatible providers by registering @opencode-ai/openai-compatible as a bundled provider. It introduces two new configuration options: omitMaxOutputTokens to exclude the max_output_tokens parameter for incompatible proxies, and useResponsesApi to control whether the Responses API or Chat Completions API is used.
Key changes:
- Register
@opencode-ai/openai-compatibleprovider with support for both Chat and Responses APIs - Add
omitMaxOutputTokensoption to conditionally omitmax_output_tokensparameter - Default to Responses API with fallback to Chat API via
useResponsesApi: false
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.
| File | Description |
|---|---|
packages/opencode/src/provider/sdk/openai-compatible/src/responses/openai-responses-language-model.ts |
Modified to conditionally include max_output_tokens parameter based on omitMaxOutputTokens config option |
packages/opencode/src/provider/sdk/openai-compatible/src/responses/openai-config.ts |
Added omitMaxOutputTokens field to OpenAIConfig type; removed documentation for fileIdPrefixes |
packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts |
Added omitMaxOutputTokens option to provider settings and passed it through to responses model; removed documentation for existing fields |
packages/opencode/src/provider/provider.ts |
Registered @opencode-ai/openai-compatible as bundled provider and added API selection logic based on useResponsesApi option |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
packages/opencode/src/provider/sdk/openai-compatible/src/responses/openai-config.ts
Show resolved
Hide resolved
packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts
Show resolved
Hide resolved
| } else if (model.api.npm === "@opencode-ai/openai-compatible") { | ||
| language = | ||
| provider.options?.useResponsesApi === false | ||
| ? (sdk as any).chat(model.api.id) | ||
| : (sdk as any).responses(model.api.id) | ||
| } else { | ||
| language = sdk.languageModel(model.api.id) | ||
| } |
There was a problem hiding this comment.
The new API selection logic for @opencode-ai/openai-compatible providers lacks test coverage. The conditional logic that determines whether to use sdk.chat() or sdk.responses() based on the useResponsesApi option should be tested. Consider adding tests that verify: 1) the default behavior uses responses(), 2) setting useResponsesApi to false uses chat(), and 3) the omitMaxOutputTokens option is properly passed through to the underlying model.
| * - OpenAI: ['file-'] for IDs like 'file-abc123' | ||
| * - Azure OpenAI: ['assistant-'] for IDs like 'assistant-abc123' | ||
| */ | ||
| fileIdPrefixes?: readonly string[] |
There was a problem hiding this comment.
The newly added omitMaxOutputTokens field lacks documentation explaining its purpose and usage. Consider adding a JSDoc comment to describe when and why this option should be used, similar to the documentation that was removed for the fileIdPrefixes field.
| fileIdPrefixes?: readonly string[] | |
| fileIdPrefixes?: readonly string[] | |
| /** | |
| * If true, do not send a `max_output_tokens` (or equivalent) parameter to the | |
| * OpenAI-compatible API, even when a max output value is available. | |
| * | |
| * Use this when targeting backends that either do not support a | |
| * `max_output_tokens` setting or have their own fixed token limits, to avoid | |
| * request validation errors or conflicting limits. | |
| */ |
packages/opencode/src/provider/sdk/openai-compatible/src/openai-compatible-provider.ts
Show resolved
Hide resolved
GreyElaina
force-pushed
the
feat/openai-compatible-responses-api
branch
from
b68ccc1 to
acfc895
Compare
1 participant
Summary
@opencode-ai/openai-compatibleas bundled provider with Responses API supportomitMaxOutputTokensoption for providers that don't supportmax_output_tokensparameterresponses()API, with option to fallback tochat()viauseResponsesApi: falseMotivation
Many OpenAI-compatible API proxies (OpenRouter, NewAPI? etc.) now support OpenAI's
/responsesendpoint. However,@ai-sdk/openai-compatibleonly supports Chat Completions API.This PR exposes the existing internal Responses API implementation (currently only used for GitHub Copilot) to custom providers.
Related: vercel/ai#9723 (stale, 795 commits behind)
Usage
{ "provider": { "my-provider": { "npm": "@opencode-ai/openai-compatible", "api": "https://my-api.com/v1", "env": ["MY_API_KEY"], "options": { "omitMaxOutputTokens": true }, "models": { "gpt-5.2": {} } } } }Options
omitMaxOutputTokensmax_output_tokensparameter (some proxies don't support it)useResponsesApifalseto use Chat Completions API