Is OrchKernel an agent framework?

No. Frameworks such as LangChain and CrewAI build agents. OrchKernel governs them and runs alongside whatever built them. It also ships ready-made agents by department, and you can build your own as records with a plain-language playbook.

Do we have to rebuild the agents we already use?

No. Outside agents such as Claude, Cursor or your own code get their own identity and act only through the gate. They connect through the governed actions API, its TypeScript and Python clients, or the MCP gateway.

What happens when an agent tries something risky?

The gate answers allowed, needs approval or denied, with the reason. For example, a $240 refund waits for an admin, then runs exactly once.

Can we prove what an agent did?

Yes. Every event is chained to the one before it, and ok audit verify names the first break. Any run can be replayed step by step.

Where does our data go?

OrchKernel is self-hosted. You choose your model providers, and each model is cleared for a data class, so restricted data reaches only models cleared for it.

Why OrchKernel

Frameworks build agents. OrchKernel governs them.

It runs alongside whatever built them.

Become a design partner See how it works

Why now

Agents now act

Agents have moved from answering questions, to suggesting inside one tool, to acting. They email customers, issue refunds, change records and ship code.

Once an agent can act, every company needs two answers. Who is it acting for, and who said it could?

Where it fits

It governs agents, whatever built them

OrchKernel is the governed layer where people, agents and systems work under one set of rules.

Not an agent framework

LangChain and CrewAI build agents. OrchKernel governs them, whatever built them.

Not a workflow builder

Tools like n8n chain steps together. OrchKernel decides whether an action may run, and who must approve it.

Not a chatbot platform

Its agents act inside your systems: they email customers, issue refunds and change records, under your rules.

Not a prompt tool

An agent is a record: a playbook, declared tools, a trigger, a trust tier, a budget and a person it acts for.

The rule underneath everything

Agents propose. Rules decide. The kernel applies.

support-agentsales-agentfinance-agentAgents, whoever built them

the gateOne gate for every action

ZendeskStripeGitHubThe systems they act on

PASS allowed by a rule·HOLD waits for a person·DENY stopped

Agents propose

An agent asks before it acts. The gate checks kill switches, arguments, delegation, budgets, rules, then risk against trust.

Rules decide

Routine actions are approved automatically. When one needs review, a reviewing agent or the right person sees the exact payload. An agent never has more than the person it acts for.