The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Recordant listing page.
Recordant is a patent-pending diagnostic + decisions-ledger engine for any host software system. Point it at a target's repo, changelog, and logs, and it runs a periodic scan that produces an operator brief: what changed, what looks broken, what was fixed, and a prioritized diagnosis list — plus a chat-native MCP console over the same data.
Bring your own LLM (Anthropic, OpenAI, or Gemini — your key, your usage, your bill). Recordant makes no provider assumption.
Install the package and your LLM provider's SDK (only the one you use):
Copy recordant.example.toml to recordant.toml and set:
repo_root — the target repo to diagnosechangelog_path — usually CHANGELOG.md (falls back to git log if absent)log_globs — which log files to health-scan[llm] — your provider, model, and api_key_env (the name of the
environment variable holding your key — never the key itself)Export your key and run one cycle:
The brief path prints to stdout. The brief, snapshot, and (optional)
decisions ledger are written under your configured output_dir.
Schedule it (cron / systemd timer / your own scheduler) at the
schedule_seconds cadence in your config.
That's the whole loop. See ENGINE.md for the adapter architecture, config
reference, and known limitations.
Recordant also runs as an MCP server exposing 7 tools over stdio, for use from a chat-native MCP client:
| Tool | Purpose |
|---|---|
recordant_brief | Latest delta brief since the prior cycle |
recordant_status | Current status snapshot (health, cadence, active book) |
recordant_digest | Open items, Accepted-Risk Register, fixes shipped, deadlines |
recordant_open_items | Every open diagnosis with age, recurrence, evidence |
recordant_acknowledge | Acknowledge a diagnosis (P0 items cannot be acknowledged) |
recordant_accept_risk | Pin a diagnosis in the Accepted-Risk Register |
recordant_resolve | Assert a diagnosis is resolved (audited against the changelog) |
Token-auth is on by default — there is no anonymous mode. The server
requires a valid RECORDANT_LICENSE_KEY and verifies it against the license
service at startup before registering a single tool; it will not run
unlicensed, even briefly. A short offline-grace window covers a transient
network failure to the license service, but an explicitly invalid or expired
key is refused immediately.
Add it to your MCP client's config (Claude Desktop, Claude Code, etc.):
Recordant is licensed software, not open source (see LICENSE). Get a key
and see current pricing at
https://mentorsentinel.ai/markets-ai-governance.html.
Recordant ships as a compiled image -- there is no engine source to build
from, so docker build is not part of a customer workflow. Pull the
licensed image (published per BUILD-RELEASE.md) and run it:
The image is signed (cosign) and ships with an SBOM (syft) so you can audit
what it does -- dependencies, egress -- without needing readable engine
source; see BUILD-RELEASE.md for verification commands.
See docker-compose.example.yml for a compose-based setup, including the
MCP console service.
Found a bug or have a feature request? Open an issue — see
.github/ISSUE_TEMPLATE/. For licensing or account questions, use the
pricing page above.
Recordant's inference loop and decisions-ledger lifecycle are patent-pending (U.S. application numbers available on request). "Patent-pending" reflects filed, unexamined provisional applications — no patent has issued.