Bring your prompt.
Leave with a better one.
Classification · Routing · Moderation · Extraction · Triage
Free first evolution. No credit card. Your prompt stays yours.
Three steps from prompt to evolved prompt.
Bring your prompt
Paste your production prompt and upload labeled examples. No data? No problem. Give us 5 examples and we'll generate 50 for you.
Watch evolution happen
We spawn 15 candidate prompts, score each against your data, and breed the top performers across up to 25 generations. Watch it happen live.
Copy your evolved prompt
Get the winning prompt with before/after accuracy and 95% CI. Copy it into your production system. Yours to deploy anywhere.
From 68% to 87% evolved.
A human-written starting prompt scored 68% on 110 real Capterra app reviews (customer-action routing). The Cambrian-evolved version hit 87% on the same data. Same model. Bred smarter through generations of selection pressure.
Across our benchmark tasks, lifts range from +6% to +34% depending on task difficulty. Average lift: +19 absolute points.
See all 11 case studies →Your prompt works on any LLM.
Evolved prompts are plain text. Paste them into Claude, GPT, Gemini, Llama, or any future model. The structural improvements (cascading rules, label definitions, tie-breakers) transfer across providers.
No vendor lock-in. Switch providers any time, your prompt travels with you.
Re-evolve when your data or model shifts.
Customer language drifts. Edge cases emerge. Model providers ship new versions. A prompt that hit 89% in Q1 can quietly slip to 80% by Q3. Same prompt, different reality.
Top teams re-evolve every 1-3 months to recapture the last 10-20% of lift portability leaves on the table.
Why not just use a built-in prompt improver?
Every LLM provider has a “make this prompt better” button. Here’s what they don’t do.
| Built-in tools | Cambrian Lab | |
|---|---|---|
| Tests on your real data | No | Yes |
| Measurable before/after accuracy | No | Yes, with 95% CI |
| Tests multiple candidates | 1 suggestion | 15 per generation |
| Iterative improvement | One pass | Up to 25 generations |
| Works across providers | Locked in | Any LLM |
| Typical accuracy gain | Marginal | +19 pts average |
Premium service. Premium results.
Every evolution runs the full 15-agent pipeline across up to 25 generations on your data.
- 5 evolutions per month
- Up to 200 training examples per run
- Prompt library + version history
- Email support
- 20 evolutions per month
- Up to 500 training examples per run
- Priority evolution queue
- Lineage downloads + shareable reports
- API access for automation
- Priority support
- Unlimited evolutions
- Private evolution infrastructure
- Regression testing + drift alerts (coming)
- Dedicated success manager
- Custom SLAs + security review
All plans: bring your own LLM API key. Inference runs on your account, we never see or store it. Most evolutions complete in 10-15 minutes and use roughly the equivalent of a small batch evaluation, not continuous usage.
Annual billing saves 2 months. Custom enterprise plans available on request.
Stop guessing. Start evolving.
Your first evolution is free. Upload your prompt, drop in 20 labeled examples, and watch a measurably better prompt emerge in under 15 minutes.
Evolve my prompt →