Skip to content

feat(models): add available_in_agent to the catalog and drop GPT-OSS and Llama from Maple Work - #1386

Merged
AnthonyRonning merged 1 commit into
masterfrom
agent/available-in-agent
Oct 30, 2026
Merged

AnthonyRonning merged 1 commit into
masterfrom
agent/available-in-agent

Conversation

@AnthonyRonning

@AnthonyRonning AnthonyRonning commented Oct 30, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Maple Work stops offering GPT-OSS 120B and Llama 3.3 70B. The server decides which catalog models Maple Work may offer, so the list can change later with a backend change only.

  • Server (model_config.rs): every model in /v1/models/catalog now carries available_in_agent. It is false for gpt-oss-120b and llama3-3-70b and true for every other listed chat model. Both models stay in the catalog and in /v1/models, so Research and the API keep them. API-only entries are set to false but never reach the catalog.
  • No Free agent default: defaults.agent is DeepSeek V4.1 Flash on every plan, so the per-plan branch is gone. Before, Free got gpt-oss-120b.
  • SDKs: optional available_in_agent on the Rust CatalogModel and on the TS ModelCatalogItem.
  • App (apps/maple-agent): selectable_catalog_models drops models with available_in_agent: false. A model without the field counts as available, because older servers don't send it. When the execution computer runs an older build, the controller applies the same rule to that computer's list.

Tasks already locked to GPT-OSS or Llama get no special handling. A saved pick of either model falls through to defaults.agent, as for any model the catalog stops offering (the order from PR #1360).

Rollout

Either order is safe:

  • The app treats a missing field as available, so it can ship first and still lists every model until the server ships.
  • Once the server ships, the two models drop out of new builds' menus.
  • Older app builds ignore the field and keep listing both models.
  • Research and other catalog readers ignore the field. Nothing outside Maple Work reads defaults.agent; I checked apps/maple-research, sdk, proxy and the docs.

Validation

  • services/opensecret: fmt, clippy --all-targets --all-features -D warnings, cargo test --all-features: 1078 passed, 54 ignored. Two new tests:
    • defaults.agent names a listed, enabled, tool-capable model with available_in_agent: true on every plan.
    • Every catalog model emits the field, and GPT-OSS and Llama are still listed in the catalog and in /v1/models.
  • sdk/rust: fmt, clippy -D warnings, lib tests (123 passed), cargo doc -D warnings.
  • sdk (TS): prettier, build (tsc + vite), unit tests (259 pass).
  • apps/maple-agent, through the pre-commit hook:
    • just ci-lint: fmt, clippy -D warnings for every feature set, packaging tests.
    • just ci-test: maple-agent 781 passed, app 1150 passed, Pi crates all pass.
    • New unit test: available_in_agent true, false and missing.
    • just ci-extra: every target builds, headless tests 352 passed.
  • Local mde stack on this branch (OpenSecret, billing, Continuum):
    • Signed in as the Free and Pro fixtures and fetched the catalog over the encrypted API with the in-tree Rust SDK.
    • Drove the real app with cua-driver (screenshots below).

Catalog from the local server on this branch. Excerpt: one line per model, other fields left out. Free and Pro match.

{
  "data": [
    {"id": "gpt-oss-120b", "enabled": true, "available_in_agent": false},
    {"id": "gemma4-31b", "enabled": true, "available_in_agent": true},
    {"id": "glm-5-3-flash", "enabled": true, "available_in_agent": true},
    {"id": "kimi-k3", "enabled": true, "available_in_agent": true},
    {"id": "glm-5-3", "enabled": true, "available_in_agent": true},
    {"id": "deepseek-v4-1-flash", "enabled": true, "available_in_agent": true},
    {"id": "deepseek-v4-1-flash-1017-uncensored", "enabled": true, "available_in_agent": true},
    {"id": "llama3-3-70b", "enabled": true, "available_in_agent": false}
  ],
  "defaults": {"quick": "auto:quick", "powerful": "auto:powerful", "agent": "deepseek-v4-1-flash"}
}

On master's server, the same request returns no available_in_agent on any model, and Free's defaults.agent is gpt-oss-120b.

Screenshots

New task, Pro fixture, server on this branch: GPT-OSS and Llama 3.3 are gone.

Maple Work model menu without GPT-OSS and Llama 3.3

Same app build, server running master's model config (no available_in_agent): the app treats the missing field as available and lists all eight models. This is also the before state.

Maple Work model menu listing GPT-OSS and Llama 3.3 against a server without the field

🤖 Generated with Claude Code


View with [code]smith Autofix with [code]smith
Need help on this PR? Tag @codesmith-bot with what you need.

…and Llama from Maple Work

Every model in /v1/models/catalog now carries available_in_agent.
GPT-OSS 120B and Llama 3.3 70B are false, so Maple Work stops offering
them; the catalog and /v1/models still list both for Research and the
API. defaults.agent is DeepSeek V4.1 Flash on every plan, so Free no
longer gets GPT-OSS.

The Rust and TypeScript SDK catalog types gain the optional field.
Maple Work drops models marked false and treats a missing field as
available, so either side can deploy first.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@AnthonyRonning
AnthonyRonning marked this pull request as ready for review October 30, 2026 17:29
@AnthonyRonning
AnthonyRonning merged commit 7e6674d into master Oct 30, 2026
28 checks passed
@AnthonyRonning
AnthonyRonning deleted the agent/available-in-agent branch October 30, 2026 17:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant