Karjan research preview

Agents, tools and sandboxes, managed for you.

Define agents in YAML, write tools in TypeScript, and let Karjan run them: spin up 100 agents for a batch job in minutes, or keep a small standing system running on schedules, webhooks and messages.

Isolated projects
Every project is its own world: agents, sandboxes, drives, secrets and code never reach another project.
Fine-grained access
Give each agent exactly the tools and sandboxes it needs. Scope each secret to exactly the tools that use it.
Batches and standing systems
Fan out one agent over a thousand inputs, or run persistent agents woken by schedules, webhooks and messages.
Files are the source of truth
Edit in the UI or locally with the CLI. Coding agents can work on your project like on any codebase.
Your own LLM keys
Inference runs on your OpenRouter, Anthropic or OpenAI key. Secrets never enter the model's context.
No surprise bills
Per-turn, per-project and per-account limits, a pause-everything button and live cost estimates.