Run LangChain agents on a governed runtime — Kohala
Run your LangChain chains and agents on Kohala's governed runtime. Import your code — it lands under a recorded charter, a server-enforced allowlist on the Kohala tools it can call, and a billing cap, unscheduled unless you ask for a cadence — or wrap your AgentExecutor with the kohala SDK and report each run, capped, validated, metered, and traced.
Explore Kohala
Frequently Asked Questions
- Can I run my existing LangChain agent on Kohala?
- Yes, two ways. Paste your LangChain code into the Kohala import wizard and your loop becomes the agent's scripts on the governed runtime, under a recorded charter, a server-enforced allowlist on the Kohala tools it can call, and a billing cap. Or install the kohala Python SDK, keep the AgentExecutor on your own infrastructure, and report each run — Kohala meters it against the same caps, runs your output validators, and records the trace. Neither path rewrites the agent.
- Do I have to change my LangChain code?
- No. The import path runs your code as-is on Kohala's runtime, and the connect path is a one-line wrap around your existing executor. Your chains, tools, and prompts stay exactly as you wrote them.
- What does a governed run add to LangChain?
- On the hosted runtime every Kohala tool a run calls is admitted against the agent's server-enforced allowlist, and the run is recorded under its charter, metered against a billing cap, validated on output, and recorded with a trace ID, latency, and credit cost. A run you report from your own infrastructure is recorded under the same charter and held to the same caps and validators, with the same metering and trace — its tool calls happen in your process, so nothing admits them. Account-scoped shared metered memory is available for the agent to read and write once you wire it in — it is not something the import arrives with.