EKKA DEVELOPER GUIDE
The governed runtime for AI agents¶
An agent is the AI equivalent of a human expert. It holds plans: the things that expert knows how to do, in the order they belong in. You write the plans. EKKA runs them in your own environment and proves what happened.
Every action an agent takes is authorized before it happens and signed after. Anyone can check the record for themselves, offline.
EKKA is the governed runtime where AI agents become software you can trust. Experts encode what they know as execution plans anyone can run. Gates enforce, Govern decides and never sees your credentials, and receipts prove what happened and meter what it cost. The category is governed execution.
Governance is built into how your agent runs. Every allowed action comes back with a signed receipt you can verify offline. Anything without a Grant stops before any I/O, and that decision is recorded too.
Your prompts, your rows, your files and your credentials never reach EKKA. We can prove what happened because we signed it, not because we saw it. That is the security model, and every claim in it comes with the command that checks it.
Get started¶
-
Five acts, about fifteen minutes: get a runtime you control, meet your agent, give it a job, point it at your own API, then verify the proof.
-
Multi-step plans and schedules →
Several governed steps in order, then the same plan on a schedule. A timer is not a less-governed path.
-
The four words that are EKKA: Gate, Grant, Gate Token, and Receipt, plus agents, plans, and the Enclave, what you build and how EKKA governs it.
Governance and proof¶
-
The principles your organization writes down to govern its AI agents, kept immutable and versioned with a fingerprint anyone can check. You write the words. EKKA supplies the courtroom.
-
A rule EKKA checks before a plan runs: office hours, a spend ceiling, a rate limit. Every rule watches before it blocks, so you see what it would have stopped before it stops anything.
-
One operation, slowed right down: refused with no Grant, authorized, allowed, and returned as signed proof.
-
Nothing moves without permission, and every claim on the page ships with the command that proves it. Grants, deny before I/O, what EKKA can never see, and proof you verify offline.
-
What the signed audit chain proves, how to read it, and how to verify it yourself, offline.
-
Field by field: what EKKA receives, what it never receives, and why there is nothing on our side to train on.
Build with EKKA¶
-
A folder on your own disk that is the agent's memory, a plan that adds to it on a schedule, and a weekly memo written with a prompt EKKA never reads. Our flagship worked example.
-
Work written once and called from any plan in one step. A version is never rewritten, and nothing you write reaches your organization until a person accepts it.
-
Any HTTP API, one operation at a time. Slack, Sentry and Jira are in the catalog; your own goes in beside them.
Run it in production¶
-
An Enclave that keeps working when the laptop is closed.
-
Every capability is on from day one, bar one that is a separation-of-duties boundary. What your agent can reach, what EKKA counts, and the daily ceiling that caps your bill.
-
$5 of credit to start, no card. After that, one cent per governed operation, and a spend ceiling you can see and change.
-
Every command you'll use, from
ekka logintoekka receipts verify.
Private beta
EKKA is in private beta. Write to contact@ekka.ai to request access.
Evaluating EKKA for a team?
This site is written for people building agents. The governance model, the trust story and how EKKA fits an organization live on ekka.ai.