Skip to content

feat: add Fireworks AI provider and fix model tier resolution#670

Merged
FL4TLiN3 merged 5 commits intomainfrom
feat/fireworks-provider
Mar 3, 2026
Merged

feat: add Fireworks AI provider and fix model tier resolution#670
FL4TLiN3 merged 5 commits intomainfrom
feat/fireworks-provider

Conversation

@FL4TLiN3
Copy link
Contributor

@FL4TLiN3 FL4TLiN3 commented Mar 2, 2026

Summary

  • Add Fireworks AI as a new LLM provider (@perstack/fireworks-provider)
  • Fix model tier resolution to cascade high → middle → low instead of hardcoded "middle" fallback

Changes

Fireworks provider

  • New package: packages/providers/fireworks/ — adapter, error handling, unit tests
  • Core schemas: Added "fireworks" to ProviderName, FireworksProviderConfig
  • Known models: Kimi K2.5, DeepSeek V3P2, GLM-5 (all high tier)
  • Runtime: Registered in factory, register-providers, and model.ts switch
  • TUI: Added FIREWORKS_API_KEY env mapping and provider config builder
  • E2E: Added fireworks to multi-provider test, FIREWORKS_API_KEY to CI workflow

Model tier resolution

  • Added resolveModelTierWithFallback() that cascades high → middle → low
  • Updated coordinator-executor.ts and delegation-executor.ts to use cascade fallback
  • Fixes providers with only high tier models (e.g. Fireworks) failing when defaultModelTier doesn't match

Test plan

  • Unit tests pass (bun run test)
  • Typecheck passes (24/24 packages)
  • E2E provider tests pass locally (openai, anthropic, google, fireworks)
  • CI checks pass

🤖 Generated with Claude Code

FL4TLiN3 and others added 5 commits March 2, 2026 19:05
Add Fireworks AI as a new LLM provider, enabling access to models like
Kimi K2.5, DeepSeek V3P2, and GLM-5 via the Fireworks inference API.

- New provider package at packages/providers/fireworks/
- Register provider in core schemas, runtime factory, and TUI config
- Add known models with context window metadata
- Add E2E provider test case and CI secret for FIREWORKS_API_KEY

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The default fallback was hardcoded to "middle" tier, which caused
providers with only "high" tier models (e.g. Fireworks) to fail
model resolution. Now cascades high → middle → low via
resolveModelTierWithFallback, ensuring the best available model
is always selected.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@FL4TLiN3 FL4TLiN3 changed the title feat: add Fireworks AI provider feat: add Fireworks AI provider and fix model tier resolution Mar 3, 2026
@FL4TLiN3 FL4TLiN3 merged commit e1c24a0 into main Mar 3, 2026
11 checks passed
@FL4TLiN3 FL4TLiN3 deleted the feat/fireworks-provider branch March 3, 2026 05:01
@FL4TLiN3 FL4TLiN3 mentioned this pull request Mar 3, 2026
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