feat: add Eden AI provider (OpenAI-compatible, 600+ models, EU/GDPR)

[MVS-source]

Summary

Adds Eden AI as a native provider, mirroring the existing OpenRouter (OpenAI-compatible) path. Eden AI is an OpenAI-compatible gateway (https://api.edenai.run/v3/chat/completions, provider/model naming) giving one key + EU/GDPR-compliant access to 600+ models across OpenAI, Anthropic, Google, Mistral, etc. It exposes both OpenAI- and Anthropic-compatible endpoints, so it fits both Lynkr's OpenAI-format path and the Claude Code / Cursor / Codex passthrough, and pairs naturally with tier routing (cheap model for SIMPLE → frontier for REASONING, one key).

What changed

• Config (src/config/index.js): edenai in SUPPORTED_MODEL_PROVIDERS, config.edenai, EDENAI_API_KEY/MODEL/EMBEDDINGS_MODEL/ENDPOINT, reload path.
• Dispatch (src/clients/databricks.js): invokeEdenAI() mirroring invokeOpenRouter, reusing the shared Anthropic↔OpenAI converters; initial + fallback branches.
• Orchestrator (src/orchestrator/index.js): endpoint lookup, tool passthrough, response conversion, headroom/caveman provider sets.
• API/embeddings (src/api/openai-router.js, src/api/providers-handler.js): provider listing + /v3/embeddings parity.
• Routing/capabilities: src/routing/index.js, src/routing/model-tiers.js, src/clients/provider-capabilities.js (reasoning_content), src/clients/prompt-cache-injection.js, src/dashboard/api.js.
• UX/docs: bin/lynkr-init.js wizard entry, .env.example, documentation/providers.md.
• Tests/CI: test/edenai-integration.test.js, .github/workflows/edenai-integration.yml (EDENAI_API_KEY secret scoped to the live-test step only; self-skips on fork PRs).

15 files, +401 / −10.

Usage

MODEL_PROVIDER=edenai
EDENAI_API_KEY=your-key
EDENAI_MODEL=anthropic/claude-sonnet-4-5
# or per tier:
TIER_SIMPLE=edenai:openai/gpt-4o-mini
TIER_REASONING=edenai:anthropic/claude-opus-4-8

Testing

Live-tested against api.edenai.run/v3 with a real key:

• Chat completion (200, OpenAI shape) ✅
• Tool-calling on openai/gpt-4o, gpt-4o-mini, anthropic/claude-opus-4-8, claude-haiku-4-5 → finish_reason: tool_calls ✅
• Streaming (SSE) ✅
• node --test test/edenai-integration.test.js (config/routing/capabilities) ✅

Notes

• Implemented as an OpenAI-compatible provider; no new response-conversion code (reuses openrouter-utils).
• Reviewer: the diff intentionally follows the OpenRouter provider 1:1 for maintainability.

[MVS-source]

Hi Fast-Editor team 👋

This PR adds Eden AI as a native provider, built by mirroring your OpenRouter path 1:1 (reuses the existing Anthropic↔OpenAI converters — no new conversion code). We live-tested it against api.edenai.run/v3: chat, tool-calling (GPT-4o, Claude Opus 4.8 → finish_reason: tool_calls) and streaming.

For Lynkr users it means one EU/GDPR-compliant key → 600+ models, OpenAI- and Anthropic-compatible — so it fits both the OpenAI-format path and the Claude Code / Cursor / Codex passthrough, and pairs naturally with tier routing.

Happy to adjust to your conventions, and we're glad to maintain the provider going forward. Would love a review 🙏 — and thanks for building Lynkr!

— Eden AI