Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

[Feature]: Model-limit metadata for a Tsubasa provider entry

Open
  • Is something wrong?
#3,240 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
25/100
Issue type
Feature
Clarity
Needs clarification
Activity status
Active
Tech stack
rust, typescript
Domain
api, backend-api-design

Research direction

Start with src/crates/assembly/core/src/infrastructure/ai/provider_catalog.rs and src/shared/ai-provider-catalog/providers.json to understand overlay limits and fallback behavior. Then inspect OpenBitFun-Installer/src/data/modelProviders.ts and src/crates/adapters/ai-adapters/src/providers/openai/chat.rs. Done means the supported representation for Tsubasa's limits and capabilities is decided, with compatibility and request/response acceptance checks identified.

Written by the indexing model from the issue text.

Description

Problem / opportunity

Tsubasa exposes two public Chat Completions model aliases, tsubasa-fast and tsubasa-pro. A named entry could reduce manual endpoint/model setup across OpenBitFun's provider catalog and Installer, but a bare provider overlay would not carry their 32,768-token context limits.

Proposed behavior

Please advise on the supported place for service-owned model limits before a provider contribution is implemented.

The proposed endpoint is https://api.tsubasa.sh/v1, using the existing openai Chat Completions adapter and a user-supplied Tsubasa API key. Fast has a maximum output of 8,192 tokens and Pro 16,384; both have a 32,768-token context window. The initial profile should expose text chat without asserting tool, vision, or reasoning support. The public aliases should remain unchanged, with no fabricated release dates or private underlying model identities.

The current overlay can list model IDs, but ModelPolicyOverlay has no limit fields and curated_model_fallback returns limits: None. Tsubasa has no models.dev provider row to supply those facts. A record that only fills curated_models therefore does not establish correct budget metadata. Is a small, backward-compatible overlay metadata extension appropriate, or should this remain a documented manual provider until the catalog can represent those limits?

Product surface

AI provider / model adapter

Details, examples, or constraints

Inspected at a0d4437e96cc663476ac1557ea9856cae626c2e4:

Acceptance checks would cover catalog limits/capabilities, existing configuration compatibility, Installer/Web UI selection, exact Chat Completions request serialization and streamed response decoding, and rejected/missing credentials. An actual request should pass a 32K total-context check before broader agent/remote claims. Locale updates would follow the owning surfaces rather than sharing Web UI catalogs with the Installer.

This AI-assisted proposal follows the early design discussion requested by CONTRIBUTING.md. It contains no implementation or runtime test claim. Hosted inference and agent tool behavior have not been qualified; the current public model-discovery check returned HTTP 404.

Dominant language
Rust
Stars
2.3k
Forks
236
Avg merge
2h 39m
Merged PRs (30d)
515

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from GCWing/OpenBitFun

All issues in GCWing/OpenBitFun

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.