Open source · Apache 2.0 · v0.1 early, working, tests passing
Every AI agent needs a logbook. Yours should stay in Europe.
An open-source, tamper-evident logbook for AI agents. It records the actions, tool calls and human approvals your agent logs, on your own server. Works with any model.
First entry in five minutes: install with pip, point your agent at the server, run logbook verify.
Hosted version: 30-day free trial, cancel any time. If it is not live by the end of your trial, we cancel it rather than charge you.
Tamper-evident
Every entry carries the fingerprint of the one before it. Change, insert, remove or reorder a line and the verify command tells you exactly where. The whole mechanism is one file of about 80 lines you can read.
Any model, any cloud, or no cloud
Built on MCP, the open standard for connecting agents to tools, so Mistral, GPT-NL, Llama, Claude or GPT all write to the same log. One Docker container on your own server. No outbound network, proven by a public test.
Audit-ready
Who, what, why, approved by whom, with what outcome. Secrets masked before writing. Exports to JSON or CSV. Your DPO or auditor reads it without the tool. Maps to the EU AI Act duties for record-keeping and human oversight.
Why this exists
The EU AI Act requires providers and deployers of high-risk AI systems to keep logs and keep a human able to intervene. Those duties were due on 2 August 2026, or 2 August 2027 for AI built into regulated products. The Digital Omnibus on AI moved them to 2 December 2027 for most high-risk systems and 2 August 2028 for AI built into regulated products. The clock moved; it did not stop. Reporting serious incidents sits in a different part of the Act, outside that postponement; check its timing for your system. The logbook gives you a verified timeline of the agent's logged actions to report from.
Autonomous agents make those duties harder. One task is ten tool calls. When something goes wrong, nobody can reconstruct which call did what, on whose approval.
Most tools for this live inside a US cloud platform or are built for one model vendor. If you run a sovereign setup, an open-weight model on your own hardware, we found nothing small, neutral and tamper-evident to drop in. This is it: a tamper-evident flight recorder for AI agents that stays on your own servers.
Sources: Regulation (EU) 2024/1689 (AI Act), Articles 12, 14, 26(6) and 73; Regulation (EU) 2026/1744 (Digital Omnibus on AI), in force 27 July 2026, amending Article 113 of the AI Act.
No black box
Every claim on this page has something you can check yourself, without asking us.
Tests run in the open on every change
CI badge and every run on GitHub
No outbound network, proven
tests/test_no_network.py blocks outbound socket connections in the Python runtime and still completes a full write, verify and export cycle
We say what it does not protect against
THREAT-MODEL.md
We say which AI Act articles it helps with, and which it does not
docs/ai-act-mapping.md
The log format outlives the software
SPEC.md, Apache 2.0, separate from the code
Security issues have a process
SECURITY.md: security@delfen.ai, acknowledged within 3 working days
The public roadmap lists what v0.1 does not do yet, and the ideas we rejected.
Follow the build
Get the release notes, the v0.2 design for anchoring the log to an external record, and the first results from a regulated environment. A few mails a year, plus the weekly Delfen CTO update. Unsubscribe any time.
Free. One click to unsubscribe.
Apache 2.0 licence. Built in the Netherlands by Delfen.