feat(opencode): auto-discover models from OpenAI-compatible providers by sethjones · Pull Request #32731 · anomalyco/opencode

sethjones · 2026-06-17T20:28:35Z

Issue for this PR

Type of change

New feature

What does this PR do?

Auto-discovers available models from OpenAI-compatible providers that have a baseURL configured in opencode.json. Instead of requiring users to list every model manually, opencode now calls GET /v1/models at startup and populates the model list.

Models come from three sources: Models.dev (metadata), config file (overrides), and the provider API (actual availability). A model is visible only if the provider API returns it or it is declared in config. Metadata is merged with API > config > Models.dev precedence.

This means providers like LM Studio, Ollama, and self-hosted OpenAI-compatible servers automatically show their loaded models without any per-model configuration.

New in this iteration:

discoverModels: false opt-out — Set discoverModels: false at the provider top level to skip discovery entirely. Useful when you want to manage models exclusively through config or Models.dev.

{
  "provider": {
    "my-local": {
      "npm": "@ai-sdk/openai-compatible",
      "discoverModels": false,
      "options": { "baseURL": "http://localhost:8080/v1" }
    }
  }
}

Context & output limits from API — Discovery now parses context_length (or max_context_length) and max_output_tokens from the provider API response and applies them to discovered models. Config-defined limits always take precedence.

Warning logs — Discovery failures (timeout, network error, non-2xx status, JSON parse failure) now emit Effect.logWarning with provider ID and reason. Additionally, when config-defined limits exceed the discovered limits, a warning is logged so users know their config may cause runtime failures.

How did you verify your code works?

Built a standalone binary and ran it with a llama.cpp server configured via baseURL
Confirmed models appear in the TUI /models list without manual per-model config
Confirmed models not returned by the provider API are hidden, even if listed in Models.dev
Confirmed startup completes without hanging or displaying discovery output to the terminal
Ran bun typecheck from packages/opencode — passes
Ran provider tests — 97 tests pass

Checklist

I have tested my changes locally
I have not included unrelated changes in this PR

…oviders

- Fetch models from OpenAI-compatible providers with baseURL in config - Support plugin-registered discovery loaders (e.g., GitLab) - Discovery runs synchronously during bootstrap init - Remove providers with no models after successful discovery - Add tests for discovery and model registration

- Extract discoverProviderModels() helper from inline 110-line Effect.gen - Replace globalThis.fetch with HttpClient layer (FetchHttpClient) - Add Effect.timeout(10s) to HTTP discovery requests - Replace Exit._tag with Exit.isSuccess/IsFailure - Remove redundant config.get() in init() (use captured s.cfg) - Remove unnecessary 'as any' casts - Fix indentation to consistent 2-space - Set family: 'discovered' on discovered models - Update discovery tests to use HttpClient layer mocks

github-actions · 2026-06-17T20:30:09Z

The following comment was made by an LLM, it may be inaccurate:

Potential Duplicate PRs Found

Based on my search, I found several related PRs that may be addressing similar or overlapping functionality:

PR feat(opencode): local LAN provider discovery + auto-discover models #27554 - feat(opencode): local LAN provider discovery + auto-discover models
- Appears to address local LAN provider discovery combined with auto-discovery of models
PR feat(core): add OmniRoute custom provider support #32170 - feat(core): add OmniRoute custom provider support
- May be related to custom provider configuration and model discovery
PR feat(opencode): add local server provider with auto model discovery #19959 - feat(opencode): add local server provider with auto model discovery
- Directly addresses auto model discovery for local servers
PR feat(provider): add Requesty model discovery from /v1/models #28973 - feat(provider): add Requesty model discovery from /v1/models
- Specifically implements /v1/models endpoint discovery (same approach as current PR)
PR feat(provider): add endpoint-based custom providers and model discovery #29392 - feat(provider): add endpoint-based custom providers and model discovery
- Addresses endpoint-based custom providers with model discovery functionality

These PRs should be reviewed to determine if they:

Are already closed/merged solutions to the same issue
Address overlapping functionality that needs to be consolidated
Have different scopes that complement each other

I recommend checking PR #27554, #19959, #28973, and #29392 in particular, as they appear most directly related to auto-discovering models from providers.

…vered models

… exceeding discovered

sethjones · 2026-06-18T00:04:47Z

The following comment was made by an LLM, it may be inaccurate:

Potential Duplicate PRs Found

Based on my search, I found several related PRs that may be addressing similar or overlapping functionality:

1. **PR [feat(opencode): local LAN provider discovery + auto-discover models #27554](https://github.com/anomalyco/opencode/pull/27554)** - `feat(opencode): local LAN provider discovery + auto-discover models`
   
   * Appears to address local LAN provider discovery combined with auto-discovery of models

2. **PR [feat(core): add OmniRoute custom provider support #32170](https://github.com/anomalyco/opencode/pull/32170)** - `feat(core): add OmniRoute custom provider support`
   
   * May be related to custom provider configuration and model discovery

3. **PR [feat(opencode): add local server provider with auto model discovery #19959](https://github.com/anomalyco/opencode/pull/19959)** - `feat(opencode): add local server provider with auto model discovery`
   
   * Directly addresses auto model discovery for local servers

4. **PR [feat(provider): add Requesty model discovery from /v1/models #28973](https://github.com/anomalyco/opencode/pull/28973)** - `feat(provider): add Requesty model discovery from /v1/models`
   
   * Specifically implements `/v1/models` endpoint discovery (same approach as current PR)

5. **PR [feat(provider): add endpoint-based custom providers and model discovery #29392](https://github.com/anomalyco/opencode/pull/29392)** - `feat(provider): add endpoint-based custom providers and model discovery`
   
   * Addresses endpoint-based custom providers with model discovery functionality

These PRs should be reviewed to determine if they:

* Are already closed/merged solutions to the same issue

* Address overlapping functionality that needs to be consolidated

* Have different scopes that complement each other

I recommend checking PR #27554, #19959, #28973, and #29392 in particular, as they appear most directly related to auto-discovering models from providers.

Here's how this PR differs from each referenced PR:

#27554 (#27554) (androidand) — Solves a broader problem: LAN-level provider discovery (mDNS, subnet scanning, TCP probing) plus model discovery. This PR is scoped specifically to config-file-driven model discovery for already-configured providers, using the Effect HttpClient layer instead of raw Promise.all/fetch. The discoverModels: false opt-out and warning logs for config-vs-discovered limit mismatches are not present in that PR.

#32170 (#32170) (MohammadMD1383) — Adds a single hardcoded OmniRoute provider with its own auth plugin. It doesn't apply to arbitrary user-configured OpenAI-compatible endpoints, doesn't parse context/output limits, and has no opt-out mechanism.

#19959 (#19959) (hmblair) — A ~50-line proof-of-concept for a single hardcoded "local" provider with an autoload boolean return. Doesn't support multiple providers, doesn't parse limits, doesn't filter Models.dev, and has no opt-out or warning logs.

#28973 (#28973) (Thibaultjaigu) — Specific to Requesty only (hardcoded URL). Solves the user-specific routing policies problem, not general OpenAI-compatible endpoint discovery.

#29392 (#29392) (dasdkslkd) — Closest in concept but orthogonal: adds a runtime UI flow for adding/removing custom providers through the app. Doesn't parse limits, doesn't filter Models.dev, has no opt-out, and doesn't use the Effect HttpClient layer.

This PR's differentiating aspects are:

Works for any provider configured with baseURL in opencode.json (not hardcoded to a single provider)
Discovers context_length and max_output_tokens from the API
Filters out Models.dev models not actually returned by the provider
discoverModels: false per-provider opt-out
Warning logs for discovery failures and config limits exceeding discovered limits
Uses the Effect HttpClient layer (FetchHttpClient) with 10s timeout (10s might be too long. I've considered 5s, but would love some feedback.)
Atomic result application via init() instead of mutation during layer construction

…nai-compatible schemas - Log warning and continue when provider disappears between discovery snapshot and apply - Add @ai-sdk/openai-compatible to schema sanitization check - Add test case for openai-compatible schema sanitization

Thibaultjaigu · 2026-06-23T18:06:46Z

@sethjones anyway we can add this?

We have now had over 445 users complain about this issue.

sethjones · 2026-06-23T18:19:59Z

@sethjones anyway we can add this?

We have now had over 445 users complain about this issue.

I'm not sure how to get the team to prioritize / respond versus what I've done already with making sure it's ready and linked.

…anomalyco#32731) Merge PR anomalyco#32731 by sethjones into dev branch. Auto-discovers available models from OpenAI-compatible providers that have a baseURL configured. opencode now calls GET /v1/models at startup and populates the model list from the provider API. Features: - discoverModels: false opt-out per-provider - Parses context_length and max_output_tokens from API - Warning logs for discovery failures and config limits exceeding discovered limits

…roviders (anomalyco#32731)" This reverts commit aad7116, reversing changes made to dfeb1b5.

…viders A configured provider can opt in (discover: true, or the v1 config's discoverModels: true) to have opencode call GET {baseURL}/models and add whatever the endpoint reports to the catalog, instead of requiring every model to be hand-listed -- useful for LM Studio, vLLM, llama.cpp/llama-swap, or any generic OpenAI-compatible self-hosted endpoint. Ported from upstream anomalyco#6231 (247 reactions/57 comments -- the second-highest reaction count found across a full survey of this repo's open issues). Three substantial PRs sat open and unmerged for 3+ months despite that reaction count. PR anomalyco#32731 (8 files, tightly scoped) is the primary source: this fork reuses its discoverModels field name, which limit fields to read from the response, the discovered-model shape, and the "configured/models.dev wins" merge rule. Deliberately changed from anomalyco#32731: discovery is opt-in per provider here, not automatic for every provider with a baseURL -- always-on would fire a startup network request at every existing custom provider. Unknown limits are left at 0 rather than anomalyco#32731's guessed 128k default. PR anomalyco#42660's URL-with-existing-path handling was reused, rewritten with this repo's own HttpClient/Schema idioms instead of raw fetch in a try/catch; its UI and endpoint weren't ported. PR anomalyco#27554 (100+ files) had nothing usable beyond confirming scope overlap. New, since none of the three PRs target it: a v2 Catalog.Service plugin (this repo didn't have a plugin/catalog system when those PRs were written). Runs in both of this repo's provider surfaces: the legacy provider service (inline before the filter pass, what the TUI model picker reads) and a new v2 Catalog plugin (packages/core/src/config/plugin/ provider-discovery.ts, once at startup then hourly, same schedule as the models.dev refresh). Known gaps: the legacy provider service has no periodic refresh (a newly-loaded local model appears only after restart/reload); stale models aren't hidden; discovered capabilities are guesses (no vision/reasoning detection); only the standard OpenAI models response shape is understood, not Ollama's native /api/tags; no settings UI or HTTP endpoint; legacy auth only checks a stored credential or options.apiKey, not plugin OAuth loaders. New packages/core/test/provider-discovery.test.ts + config/ provider-discovery.test.ts (unit-level) and packages/opencode/test/ provider/discovery.test.ts (against a real local Bun.serve server, not a mock). Verified against this fork's current dev, not just the agent's own stale worktree base: bun typecheck clean across all 30 packages; packages/core 1187 pass (only the pre-existing pty flake); packages/opencode 3617 pass, only the 4 pre-existing cf-ai-gateway failures. bun run generate from packages/client correctly produced no changes -- this is a config-only field, not a Protocol/HttpApi schema change. Co-Authored-By: Claude Sonnet 5 <[email protected]>

sethjones added 5 commits June 14, 2026 15:17

feat(provider): auto-discover models from custom OpenAI-compatible pr…

1d7ceba

…oviders

refactor: format fixes

63738c0

Merge branch 'dev' into custom-model-discovery

51dc086

sethjones added 3 commits June 17, 2026 18:14

feat(opencode): add discoverModels opt-out for provider model discovery

f257e88

feat(opencode): parse context_length and max_output_tokens from disco…

d9add28

…vered models

feat(opencode): log warnings for discovery failures and config limits…

52b7184

… exceeding discovered

sethjones marked this pull request as ready for review June 17, 2026 22:17

Merge branch 'dev' into custom-model-discovery

7d28dac

This was referenced Jun 18, 2026

📊 AI CLI 工具社区动态日报 2026-06-18 zx0828/big_model_radar#141

Open

📊 AI CLI 工具社区动态日报 2026-06-18 litang9/big_model_radar#80

Open

feat(core): support localhosting setting api link and key in connect #33103

Closed

Merge branch 'dev' into custom-model-discovery

450ac37

ProjectInitiative added a commit to ProjectInitiative/opencode that referenced this pull request Jun 28, 2026

Revert "feat(opencode): auto-discover models from OpenAI-compatible p…

686e4d6

…roviders (anomalyco#32731)" This reverts commit aad7116, reversing changes made to dfeb1b5.

github-actions Bot mentioned this pull request Jul 14, 2026

Refactors existing commits from issue https://github.com/anomalyco/op… #36790

Closed

6 tasks

This was referenced Jul 27, 2026

feat: discover Modal models #39066

Merged

Feat #6231 - automatic discovery of models from providers #39176

Closed

github-actions Bot mentioned this pull request Aug 7, 2026

feat(provider): discover local model context limits #41104

Closed

3 of 6 tasks

github-actions Bot mentioned this pull request Aug 14, 2026

feat(provider): add dynamic model discovery for custom providers #42660

Open

3 of 6 tasks

This was referenced Sep 11, 2026

feat(provider): discover models for openai-compatible providers #48486

Closed

fix(provider): discover models from openai-compatible endpoints #48510

Closed

fix(provider): discover models from openai-compatible endpoints #48549

Open

el-AI-wei mentioned this pull request Sep 14, 2026

[FEATURE]: Add LiteLLM as a provider #48938

Open

1 task done

JLor08 mentioned this pull request Sep 30, 2026

No current opencode discovers a gateway model's context size, though the docs say it reads /api/v0/models: opencode v2 asks for /api/v1/models, which 404s here, and /v1/models has no context_length JLor08/op-ai-gateway#171

Open

Navigation Menu

Repository navigation

feat(opencode): auto-discover models from OpenAI-compatible providers - #32731

feat(opencode): auto-discover models from OpenAI-compatible providers#32731
sethjones wants to merge 11 commits into
anomalyco:devfrom
sethjones:custom-model-discovery

sethjones commented Jun 17, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jun 17, 2026

Uh oh!

sethjones commented Jun 18, 2026

Potential Duplicate PRs Found

Uh oh!

Thibaultjaigu commented Jun 23, 2026

Uh oh!

sethjones commented Jun 23, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

sethjones commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Issue for this PR

Type of change

What does this PR do?

New in this iteration:

How did you verify your code works?

Checklist

Uh oh!

github-actions Bot commented Jun 17, 2026

Potential Duplicate PRs Found

Uh oh!

sethjones commented Jun 18, 2026

Potential Duplicate PRs Found

Uh oh!

Thibaultjaigu commented Jun 23, 2026

Uh oh!

sethjones commented Jun 23, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sethjones commented Jun 17, 2026 •

edited

Loading