Agenko overview
Agenko is the open-source, OTEL-native trust boundary for AI agents. One line gives every action a cryptographic receipt and a deterministic gate before any irreversible move — receipt after, gate before.
Quickstarts
Choose the path that matches where you are. Each guide is copy-paste — no setup objects.
Where to start?
From first PoC to production: instrument, receipt, gate, then operate in your own cloud.
| You need to… | Start here |
|---|---|
| Prove what the agent did | Receipts & @check |
| Stop dangerous actions | Gates & @approve |
| Run in your VPC | Self-hosting |
| Wire OpenAI or LangGraph | Integrations |
Platform
Three layers — trace and receipt what happened, gate what must not run, operate in your own stack.
Receipt & tracing
OTEL-native runs with OpenInference spans and a cryptographic receipt on every wrapped action.
Approval gates
Deterministic pause before irreversible work — human inbox, not LLM policy.
Operate & self-host
Apache 2.0 control plane — Postgres, OTLP ingest, REST API, your cloud.
Featured
Quickstart
First receipt in under a minute. Copy-paste, no config object.
Read guide →Receipts & verdicts
Claimed vs actual. Silent failures become verdicts, not green traces.
Read guide →Approval gates
Deterministic pause before irreversible work. Human inbox, not LLM policy.
Read guide →Console
Traces, receipts, and the approvals inbox in one control plane.
Read guide →How it works
- Instrument with the SDK.
agenko.init()— zero required args, never throws. Auto-detects OpenAI, Anthropic, LangChain, and more. - Receipt after every action.
@checkattaches claimed vs actual, verdict, and hash chain to the return value. - Gate before irreversible moves.
@approvepauses until a human approves in the console. - Traces land in your control plane. OTLP ingest + REST. Self-host or point at hosted cloud.
What Agenko provides
Receipt
Measured facts — claimed vs actual. Silent failures surface as verdicts, not vibes.
Gate
Deterministic block before deletes, refunds, deploys. Human inbox, not LLM policy.
Tracing
OpenTelemetry + OpenInference. Attach to the collector you already run.
Browse by topic
Full reference grouped by job — SDK primitives, integrations, operations, and open-source reuse.
Get started
- IntroductionThe trust boundary: a receipt after, a gate before.
- Agent skillContract and prompt for coding agents.
- QuickstartFirst receipt in under a minute.
- InstallationPython, TypeScript, and self-host.
- Core conceptsActions, runs, receipts, gates, verdicts.
- Best practicesWhat a good receipt and trace look like.
SDK
- init()Zero required args. Never throws.
- @checkCryptographic receipt: claimed vs actual.
- @approveDeterministic gate before irreversible actions.
- PromptsVersioned prompts. SWR cache + fallback.
- score()Langfuse-shaped measurements on a trace.
- Python SDKpip install agenko
- TypeScript SDKpnpm add @agenko/sdk
- EnvironmentAGENKO_API_KEY, AGENKO_HOST, and friends.
Integrations
The boundary
Operate
For AI agents & coding tools
Plain markdown in content/. Fetch the index, install the skill, then ground every answer in shipped SDK exports.
llms.txt
Machine index + search API docs — docs.agenko.app/llms.txt
Agent skill
Installable SKILL.md — contract, curl examples, workflow for coding agents.
Best practices
Read before instrumenting — one trace per run, honest receipts, gate scope.
Agent skill (docs)
Human-readable guide with suggested system prompt and search API usage.
Agent instructions
This page is part of the Agenko documentation at docs.agenko.app. Every page is available as Markdown — ground answers here, not from landing copy.
Querying these docs
- Full index: docs.agenko.app/llms.txt · sections: llms-docs.txt, llms-sdk.txt, llms-integrations.txt
- This page as markdown: docs.agenko.app/index.md
- Search API:
curl -sG "https://docs.agenko.app/api/search-docs" --data-urlencode "query=…"
Before writing SDK code
- Install the agent skill or read Agent skill
- Read Best practices — what a good receipt looks like
- Trust boundary:
@checkafter,@approvebefore init()— zero required args, never throws- Mark
estimated: trueuntil a check is a hard fact - Shipped vs backlog: Introduction status table