DOCS

Orkena documentation.

What the runtime does, how to deploy it, and how to produce evidence a regulator can verify offline. Everything is open to read — no account required.

Start the quickstart →

SEQ 002 · GETTING STARTED

A·01

What is Orkena

Orkena is a governed agent runtime — a runtime that executes agent workflows and enforces your controls as part of that execution, rather than reporting on them afterwards.

A·02

Concepts

The core mental model of Orkena in ~800 words: graphs, nodes, edges, runs, gates, policies, the ledger, and evidence bundles.

A·03

Quickstart — your first governed agent in 10 minutes

Step-by-step — get an invitation link, install a policy pack, run a template graph, view the ledger event, and verify the evidence bundle. Ten minutes.

A·04

Deployment options overview

Your VPC or fully air-gapped today — one Helm chart, identical features. A managed, region-pinned tier follows (2027). When to pick each, and what to expect on procurement and install.

A·05

Glossary

Every term in the Orkena vocabulary, defined. Graphs, nodes, runs, gates, policies, ledger, evidence bundles, taint, anchors, and more.

SEQ 003 · CORE CONCEPTS

B·06

Graphs and versions

How agent workflows are authored, versioned and activated in Orkena. Immutable versions, the version pin, canvas-to-code parity, and the promotion path.

B·07

Runs — durable execution

How a run is a durable state machine in Orkena: checkpoints, resume-from-checkpoint, time-travel forks, non-determinism capture and byte-level replay.

B·08

Nodes

The 19 node types in the Orkena graph DSL — agent, tool, transform, router, gates, memory and more — each with a config example.

B·09

Policy packs — how authorization works

The semantics of Orkena policy packs: authorize before execute, deny is a control not an incident, pack structure, Cedar evaluation, and how to write one.

B·10

HITL gates and four-eyes

How human-in-the-loop gates work in Orkena: quorum rules, SLA, escalation, self-approval blocks, same-team blocks, and the ledger record of every approval.

B·11

Evidence bundles

What is in an evidence bundle, how it is signed, how to verify it offline, and how to store it for retention. The artifact your auditor runs on their own machine.

B·12

The hash-chained ledger

How the Orkena audit ledger is structured — per-tenant append-only chain, hash linkage, Ed25519 anchors, and how tamper is detected.

B·12.5

Memory consolidation and gated reflection

How cross-run memory consolidation works in Orkena: overnight merging, contradiction surfacing, the human veto window, and promotion through the same eval gate as any other change.

SEQ 004 · GUIDES

C·13

Build your first policy pack

A walkthrough of authoring, testing, activating and monitoring a policy pack in Orkena — with what-if simulation before activation.

C·14

Set up four-eyes approval

Configure high-impact gates with quorum, add approver roles, enable conflict-of-interest blocks, and test the flow end to end.

C·15

Import from LangGraph

How the LangGraph adapter works — what carries over, what doesn't, and how to wrap an existing graph in policy, gates and audit.

C·16

Model provider setup (BYOK)

Configure Anthropic, OpenAI, Azure OpenAI and other model providers with scoped keys, per-model routing and bring-your-own-key.

C·17

Evaluation suites

Configure eval datasets, judge rubrics and promotion gates. No version reaches production without passing its regression suite.

C·18

Export an evidence bundle for auditors

Walkthrough of export scope, download, offline verification and hand-off to your auditor.

C·19

Air-gap deployment

Pre-flight checklist, local embedder setup, Helm install and first-run smoke test for the air-gapped deployment mode.

C·20

Incident response — kill switch

How to trigger the org, workspace, graph and agent kill switches, what each stops, and how recovery works — with the ledger record of the kill.

SEQ 005 · REFERENCE

D·21