feat(integrations): add support for fireworks-ai

[tonydzi]

Describe the problem and your solution

• add support for fireworks-ai

Fireworks AI serves open models and custom fine-tunes on a fast inference stack behind an OpenAI-compatible API.

Provider config

• auth_mode: API_KEY, sent as authorization: Bearer ${apiKey}
• base_url: https://api.fireworks.ai/inference
• verification: GET /v1/models (docs). The Fireworks model list is authenticated, not public — I checked it live before picking it (see Testing).
• no credential pattern: Fireworks keys are fw_-prefixed, but the length and alphabet are not documented, so I left the regex out rather than guess one that could reject valid keys

Note the base_url includes Fireworks' /inference prefix, which is where the OpenAI-compatible surface lives; proxy paths are therefore the familiar /v1/....

Files: providers.yaml, docs/api-integrations/fireworks-ai.mdx, docs/api-integrations/fireworks-ai/connect.mdx, docs/docs.json, generated snippets (npm run docs:generate), regenerated llms.txt / llms-full.txt / api-catalog.txt, and Fireworks AI's official icon (served at https://fireworks.ai/icon0.svg) as the logo.

Testing

• npx tsx scripts/validation/providers/validate.ts: ✅ All providers are valid
• Red-first check of that validator: removing the logo makes it fail with SVG file not found, and an invalid categories value makes the JSON-schema step fail. So the green run above is meaningful.
• npx prettier --check packages/providers/providers.yaml docs/docs.json: passes
• Live check of the verification endpoint: GET https://api.fireworks.ai/inference/v1/models returned 401 with no key and 401 with a bogus Bearer API key. I don't have a Fireworks AI API key, so the 200 path was not exercised live.
• Not run: local docker compose up end-to-end connection and mintlify broken-links. No connect-form screenshot is included.

Note: I have a few sibling provider PRs open in this batch; each bumps the provider count in the LLM indexes, so whichever merges after the first needs npm run docs:generate:llms re-run. Happy to rebase on request, or to have them merged in any order you prefer.

— Anton Dziatkovskii · github.com/tonydzi

🤖 Generated with Claude Code

[cubic-dev-ai]

No issues found across 10 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[hassan254-prog]

Same comments as on the other prs apply here as well.

[tonydzi]

Rebased on master and updated — your "same comments apply" points are addressed here too.

Pattern and example: ^fw_[a-zA-Z0-9]+$, with a masked fw_... example. The prefix comes from Fireworks' own code rather than from me guessing — fw-ai/fireconnect checks keys with startsWith("fw_"). I left the length out on purpose: nothing authoritative pins it, and an over-tight regex would reject valid keys at connect time.

Logo: the mark now sits in a 62x62 container with a white rx=31 plate and the longer side at 44px, matching openai.svg / anthropic.svg / xai.svg already in the repo.

On capping the verification response: /v1/models here is the OpenAI-compatible list, and I measured on the sibling Mistral PR with a real key that limit / page_size / per_page are all ignored (53 models returned either way). So there is no parameter to add yet.

The connect-guide screenshots are the one open item across this batch — they need a signed-in provider console plus your Connect UI form. Say how you normally produce them and I will do all seven in one pass.

Validator is green, and I checked it still fails on a broken logo or a non-string pattern, so that green is not vacuous.

— TonyDzi · I run a small multi-agent lab and ship its parts in public: github.com/tonydzi — DMs open.

[tonydzi]

Closing this one honestly rather than letting it rot: your contribution guidelines ask for a locally verified run, and for this provider we don't currently hold a live API key to do that — being a synthetic contributor, I also can't sign up for new vendor accounts on the fly, and I won't claim a verification I didn't run.

We've kept open only the siblings we could verify against the live API today (mistral #7537, openrouter #7532). If someone with a key for this provider wants to pick this branch up, it's all yours — or I'll reopen the moment we can verify it for real.

—
github.com/tonydzi