Any provider, any key
OpenAI, Groq, OpenRouter, Anthropic, and most others all work. Requests go straight from your browser to your provider—we never see the key or the bill.
AI MENTOR · FOR NOW
Bring your own key—
any provider.
Trefelle is unfunded, so there's no hosted AI mentor running on our bill yet. Instead of picking one provider for you, it runs on a key you bring—any provider works, including free ones, as long as the model you choose can reason. Vision isn't required.
WHO THIS IS FOR, HONESTLY
Right now, this is built for
the technical and the curious.
We're two people building this on the side—we don't have the funding to run a hosted AI model for everyone, which is the actual reason there's setup involved at all. Today, this is realistically for people who want to support a small, early-stage project; who are curious what a real professional workspace needs to be wired up and configured; and who are comfortable enough with tech to get an API key and paste it in.
If that's not you yet, that's completely fine—you're not the problem, the setup is. A hosted, zero-setup mentor is the actual long-term plan; this is just what's realistic before it exists.
ONE REQUIREMENT
Any key works,
if the model can reason.
Reasoning, not vision
The model just needs extended thinking / chain-of-reasoning to follow the assessment reliably. Vision isn't used, so free reasoning-only models work fine.
Stack multiple free keys
Add a few free-tier keys and Trefelle rotates to the next one automatically when one hits its rate limit—mid-conversation, without losing your place.
A FEW FREE OPTIONS
Verified against each provider's current docs—check before you commit, limits change.
OPTION 01
Groq
Reasoning models on a genuinely free tier, no card required.
- e.g.
openai/gpt-oss-120b,qwen/qwen3.x-27b - 1,000 requests/day, 200K tokens/day (free tier)
- OpenAI-compatible endpoint
OPTION 02
OpenRouter
Aggregates many providers; look for models tagged :free.
- 20 requests/minute always
- 50 requests/day (0 credits purchased)
- 1,000 requests/day (once $10+ purchased lifetime)
OPTION 03
Google Gemini
Flash-tier models have reasoning built in, on a long-standing free tier.
- e.g. Gemini Flash "thinking" models
- Google no longer publishes exact free limits—check your own AI Studio dashboard
- Has an OpenAI-compatible endpoint too
OPTION 04
Mistral (La Plateforme)
An "Experiment" free tier—use their Magistral model, their reasoning-focused line.
- Reported ~1B tokens/month cap (not from Mistral's own docs—verify yourself)
- No card, just phone verification
- Lowest-confidence pick on this list
WHAT CHANGES LATER
A hosted, zero-setup mentor
is the long-term plan.
Once Trefelle can sustainably cover inference costs, a hosted mentor will remove the setup entirely—no key, no provider choice, no rate limits to think about. Until then, bringing your own key keeps the mentor free for us to offer and honest about what it costs you.
See the approach →