Import your agent — bring LangGraph, CrewAI & LangChain agents to Kohala
Paste an agent written in LangGraph, CrewAI, PydanticAI, the OpenAI Agents SDK, or LangChain into Kai — Kohala's import wizard detects the framework, drafts the manifest, and deploys it onto the governed runtime with metering, audit trails, and schedules. No rewrite required.
Frequently Asked Questions
- Which agent frameworks can I import into Kohala?
- The import wizard detects LangGraph, CrewAI, PydanticAI, the OpenAI Agents SDK, and LangChain automatically from your imports, and also accepts plain custom Python. Each framework ships with a pre-baked dependency bundle in the sandbox, so your imports resolve without a requirements file.
- Do I have to rewrite my agent to run it on Kohala?
- No. Your framework logic stays exactly as written. The only requirement is a thin entrypoint — a run(ctx) function that Kohala calls once per shift. If your file doesn't have one, Kai can wrap your existing loop in the adapter for you and show you the diff.
- How do I import an agent with Kai?
- Open Kai, paste or upload your Python file, and say “import this agent.” Kai analyzes the source, drafts the kohala.agent.json manifest — name, charter, tool allowlist, schedule, secrets, dependencies — and waits for your approval before deploying anything. Re-importing the same name updates the agent in place instead of duplicating it.
- What does my agent gain after importing?
- The moment it lands, your agent inherits Kohala's governed runtime: a charter, a server-enforced tool allowlist, metered token billing with caps, validators, cron schedules, health tracking and alerts, a full audit trail of every shift, and the option to publish it on the marketplace.
- Can I import agents from a CI pipeline instead of chat?
- Yes. The same wizard is exposed as two REST endpoints — POST /api/agents/framework-host/analyze and POST /api/agents/framework-host/deploy — authenticated with your pk_ API key, so the import can run from a CLI or CI pipeline.
- Can agents trigger webhooks on my own infrastructure?
- Yes. Any agent can call http.post to send a payload to a URL you control — for example, to update a CRM, notify a Slack channel, or trigger a CI build. Webhook calls are metered as standard HTTP skill usage and subject to the agent's tool allowlist and billing cap.