When an agent fails a test, someone asks why. When a customer or auditor asks what the agent did last quarter, someone asks for proof. Most teams answer both questions by digging through logs, which is slow and hard to defend.
ZizkaDB now gives you two ready-made GitHub Actions and an official MCP listing that answer those questions automatically. Setup takes minutes, and each piece is open source.
The value, in plain terms
Piece The business problem What you get
ZizkaDB Why Reviewers lose hours working out why an agent test failed A causal chain in the pull request, from the failing step back to its root cause
ZizkaDB Audit Customers, auditors, and regulators ask you to prove your agent’s records are intact A hash-chain check, exported history, and a report attested by GitHub, produced on a schedule
zizkadb-mcp (MCP Registry) Developers cannot see what their editor agent did or why The same answers inside Cursor, Claude Desktop, and other MCP clients
Three outcomes follow:
Faster reviews. The cause of a failure sits in the PR comment, so the conversation moves from what happened to what to fix.
Evidence on demand. Instead of assembling proof when an auditor calls, you have a dated, verifiable report waiting. That helps with EU AI Act record-keeping and with enterprise security reviews, where a concrete answer shortens the process.
Low commitment. No account, no secrets, and no new infrastructure to start. The data stays in the runner.
Add ZizkaDB Why to your pull requests
Add this to a workflow file, such as .github/workflows/agent-tests.yml, replacing the test command with yours:
yaml
permissions:
contents: read
pull-requests: write # for the PR comment
jobs:
agent-tests:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-python@v5
with: { python-version: "3.12" }
- run: pip install -r requirements.txt zizkadb-sdk
- uses: ZIZKA-AI-SL/zizkadb-why-action@v1
with:
run: pytest tests/agent
The action starts ZizkaDB inside the runner, runs your tests, and posts the causal chain of any failure to the job summary and the PR. For this to work, your agent needs to log its steps with the SDK, linking each to the step that caused it:
python
async with ZizkaDB() as db:
ask = await db.log(agent="my-agent", event="user_message", data={"text": question})
await db.log(agent="my-agent", event="tool_call", data={"tool": "search"}, parent_id=ask.event_id)
If your agent does not log yet, the report tells you what to add. One caution: PR comments in public repositories are public, so set show-data to false if prompts are sensitive.
Add ZizkaDB Audit for recurring proof
To verify a real, running ZizkaDB each week, add a scheduled workflow:
yaml
on:
schedule: [{ cron: "0 6 * * 1" }] # every Monday
jobs:
audit:
runs-on: ubuntu-latest
permissions: { contents: read, id-token: write, attestations: write }
steps:
- uses: ZIZKA-AI-SL/zizkadb-audit-action@v1
with:
host: https://zizkadb.example.com
api-key: ${{ secrets.ZIZKADB_API_KEY }}
Each run checks every agent’s hash chain, exports the full history, and writes a report. If any event was edited, inserted, reordered, or deleted without a record, the step fails and names the event. Anyone can later verify that the report came from your workflow run, unchanged:
bash
gh attestation verify report.md -R your-org/your-repo
You can also run Audit on pull requests by giving it a run command instead of a host. That checks your logging pipeline, and the scheduled mode against your live instance is what produces evidence about real history.
Add ZizkaDB to your editor
The MCP server is listed in the official registry as io.github.ZIZKA-AI-SL/zizkadb-mcp, version 0.1.12, installed from PyPI as zizkadb-mcp. Clients that read the registry can offer it directly. To add it by hand, put this in your client’s MCP config, such as .cursor/mcp.json, pinning the version you reviewed:
json
{
"mcpServers": {
"zizkadb": {
"command": "uvx",
"args": ["zizkadb-mcp==0.1.12"],
"env": { "ZIZKADB_HOST": "http://localhost:8000" }
}
}
}
What it costs
Almost nothing to try. Both actions and the MCP server are open source, ZizkaDB self-hosts free under AGPL-3.0, and you pay only for runner minutes and the time to add two small workflow files. Managed cloud, for multiple agents, is listed at 29 euro a month for Pro and 69 euro for Team.
As an illustration, with modeled numbers you should replace with your own: if ten failing agent tests a month each take two hours to diagnose at 100 euro per hour, and the chain cuts that to 30 minutes, the saving is about 1,500 euro a month. Producing evidence for four audit requests a year, at a day or less each instead of two, saves a few thousand euro and, more importantly, turns a scramble into a routine.
What to keep in mind
Proof of integrity is not proof of correctness. The checks show records were not changed, and not that the agent was right.
Open source ZizkaDB records one agent per install. Multi-agent setups need ZizkaDB Cloud.
Pin your versions. Pin the ZizkaDB image and, for third-party actions, a full commit hash.
Mind sensitive data. Exports can contain personal data, so restrict who can download the artifacts and for how long.
Start today
Add the Why workflow to one repository and look at its next failing test. Then schedule the Audit workflow against your live instance, and run the verification command once to see the process work. The whole path takes an afternoon, and it leaves you with answers to the two questions every agent team eventually faces.
Want to test ZizkaDB on your agent? Try our open source version here: https://github.com/ZIZKA-AI-SL/ZizkaDB
Interested in a design partnership? Fill in the form on our site or reach me directly at founder@zizka.ai
Sources: ZizkaDB Why on GitHub Marketplace · ZizkaDB Audit on GitHub Marketplace · zizkadb-mcp on PyPI · ZizkaDB repository
Top comments (0)