Train, test and ship AI agents
Kojen is an AI agent training and evaluation platform. Each part works on its own and from the web app, the REST API or the MCP server.
LLM evaluations you can trust before you ship
Grade any model, prompt or deployment with an LLM judge: accuracy, hallucination, bias or your own rubric, with confidence intervals, latency and a trust score.
Learn moreAutomatic prompt optimization in the Forge
The Forge rewrites your system prompt from its failures and searches 11 hosted models for the best fit, scored on held-out test cases. No GPUs needed.
Learn moreRed-team and stress-test your agent
Attack your agent before users do: prompt injection, jailbreaks, data exfiltration, harmful content and PII leaks, scored for robustness. Plus load tests.
Learn moreLoRA fine-tuning on the cheapest GPU available
Train real LoRA adapters on Qwen3-8B, Llama-3.1-8B-Instruct or ACE-Step audio on the cheapest GPU across Modal, Runpod, Vast.ai and Lambda. Pay per use.
Learn moreDeploy as an OpenAI-compatible endpoint
Serve a tuned model as an OpenAI-compatible endpoint with streaming and tool calls, revisions, rollback, canary splits and retraining from user feedback.
Learn moreAn MCP server for training and evaluating models
Connect Claude, ChatGPT, Cursor or VS Code to Kojen over MCP and let your agent tune, evaluate, red-team and deploy models with 36 tools, signed in with OAuth.
Learn more