Keel sits between the bots you already run and the AI models you want to use — enforcing policy, capping spend, logging every decision, and holding the risky calls for a human. One layer, every process.
Your RPA does exactly what it was scripted to do, and nothing beyond it. The moment a process needs judgement, it stops or quietly breaks.
AI supplies the judgement, but you can't predict it, cap it, or prove afterwards what it did — the three things a serious process requires.
Keel is the layer in between: the judgement of AI, under the control of enterprise software.
Point your existing bots and your chosen models at Keel. Nothing is rewritten — Keel becomes the channel they talk through.
Declare budgets, permitted actions, and exactly which calls a person must approve before they run.
Every request is checked against policy, capped, and written to an audit log you can replay line by line.
Controls live in one declarative policy per agent — version it, review it, diff it. There is no hidden prompt deciding what your automation is allowed to spend or touch.
# what this agent may spend and do policy "invoice-agent" { budget = 50 EUR / day model = "any" allow = ["read_invoice", "draft_reply"] deny = ["send_payment", "delete_record"] human_review when amount > 1000 EUR log = full # replayable }
Define what an agent may and may not do. Keel checks it on every single call — not in a review afterwards, and not as a suggestion in a prompt.
Budgets per agent, team or process. A runaway loop hits the ceiling and stops there, so the month-end invoice holds no surprises.
Every decision, input and output recorded in order. When security or compliance asks what happened on the 14th, you replay it instead of guessing.
Route the calls that carry real consequence to a named person. Nothing executes until they approve it, and the approval is logged with the rest.
Keel is a licensed product, not a bespoke build. The same control layer covers every process and every team. You keep your bots, your models and your data — Keel adds the part that makes them safe to combine.
Keel is a control plane for AI agents. It sits between your existing RPA bots and AI models, adding guardrails, cost limits, full logging and human approval so automations can use AI safely in real business processes.
Enterprises and teams that already run RPA or automation and want to add AI to their processes without losing control, predictability or auditability.
Keel enforces hard budget caps per agent, team or process. It stops runaway loops before they happen, so there are no surprise bills at month-end.
Every AI decision that passes through Keel is recorded and replayable, ready for the next security or compliance review.
No. You keep your bots, your chosen models and your data. Keel is the layer they communicate through — control without a rebuild.
A monthly licence plus a fee per active agent, with optional implementation support. Customers use the software; they don't buy or own it.
Tell us what you automate today and where AI should take over the judgement. We'll show you how Keel keeps it inside the lines.
hello@keelplane.dev
Your email app should open with the message ready to send. If it doesn't, write to hello@keelplane.dev.