Skip to main content

Governance and provenance layer for agent memory.

Project description

Procheiron

A governance and provenance layer for agent memory. Not a memory engine — the trust layer that sits on top of one.

Memory engines (supermemory, mempalace, agentmemory, a vector DB, or plain files) answer "what does my agent remember?" Procheiron answers a different question: "can a different agent trust this memory, and prove who wrote it, who reviewed it, and who authorized it?" It is built for the case a single-agent memory store ignores — multiple agents sharing one governed source of truth.

Status: v0.1 draft, pre-release. Conformance passes at fixture level (a second, fully fictional deployment validates against the same Core). Honest scope discipline: nothing here claims more than the conformance suite proves.

Why this exists

Give an agent keys and it will use them. Give several agents a shared memory and, without governance, any one of them can write a "fact" the others will trust — no record of who, no independent review, no way to supersede it cleanly. Procheiron makes the trustworthy path the easy one:

  • Provenance-first lifecycle — every memory is draft → candidate → validated → active → superseded. An active (trusted) record requires independent review by an actor that is not its author.
  • Authority, not vibes — who may promote, review, or authorize is policy, checked by a validator and a policy engine, not a prompt asking nicely.
  • Provable, replayable — a conformance suite checks that a deployment obeys the spec; a fictional reference deployment (conformance/generic-vault/) proves the same Core governs a vault it has never seen.

What's in this repo

Path What it is
spec/ The v0.1 specification: governance.md, memory-commons.md, control-plane.md, the normative conformance.md MUST-list, and the Core/Profile boundary.md.
conformance/ run_conformance.py + fixtures. generic-vault/ is a complete fictional deployment ("Meridian Atelier"); minimal-vault/ is the 5-file minimal adopter. Run it to prove conformance.
examples/minimal-adopter/ The smallest compliant deployment — a memory commons with provenance + independent review, no heavyweight governance ladder.
init/ procheiron_init.py scaffolds a new deployment; PORTING_GUIDE.md is the step-by-step.

Quick start

# Install (zero runtime dependencies — stdlib-only Python):
pipx install procheiron          # or: pip install procheiron

# Scaffold a governed memory commons and validate it:
procheiron init ./my-commons
procheiron validate ./my-commons

From a source checkout (no install required):

# Prove the spec holds against the bundled fixtures:
python3 conformance/run_conformance.py

# Scaffold via the init script directly:
python3 init/procheiron_init.py --root ./my-commons

Design choices worth knowing

  • Zero runtime dependencies. Every tool a live deployment runs is standard-library Python. (jsonschema, opa etc. are used only for development/CI cross-checks, never required at runtime.) Adopt it without taking on a dependency tree.
  • Bring your own memory engine. Procheiron governs records and their lifecycle; it does not do embeddings or retrieval and never will (that is the engine's job). Point it at any store.
  • Core vs Profile. The spec is portable Core. Deployment-specific bindings (identities, paths, a git preservation executor, a full L0–L9 authority ladder) live in a profile — see boundary.md.

Not in scope (by design)

No vector/retrieval engine. No recall benchmarks — that is the memory engine's axis, not ours. No claim of "production-replicable" until a second real deployment passes conformance (fixture-level proof is what exists today, and the README says exactly that).

Roadmap (next, post first-deployment feedback)

  1. Standalone MCP servermemory.search/get/propose/promote + boot_context, the "any agent has the context on hand" integration. (A working version exists but is currently coupled to its origin deployment's layout; the port to a --root-relative standalone server is the top item.)
  2. PyPI / pipx packaging and a one-command procheiron init. Shipped in v0.1pip install procheiron, procheiron init, procheiron validate.
  3. A reference integration: Procheiron governing a third-party memory engine end to end.

License

MIT — see LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

procheiron-0.1.0.tar.gz (176.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

procheiron-0.1.0-py3-none-any.whl (75.0 kB view details)

Uploaded Python 3

File details

Details for the file procheiron-0.1.0.tar.gz.

File metadata

  • Download URL: procheiron-0.1.0.tar.gz
  • Upload date:
  • Size: 176.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for procheiron-0.1.0.tar.gz
Algorithm Hash digest
SHA256 6f0b8e5e4a1dc1a264f3c9dfdc03ae198d26d2fe8521948967eee545b650affb
MD5 a0522c63d12e8f65b2da4aa3a5a7fed0
BLAKE2b-256 4d1a1cb9b25564e74f19d5a525667ab890c3fcf4ff14f5c056ca6dcbca0a2c5b

See more details on using hashes here.

File details

Details for the file procheiron-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: procheiron-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 75.0 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.3

File hashes

Hashes for procheiron-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 2801f248348a4ee4ac4770e4822a263fc5d85547d92090cb28fd294e1fbc3435
MD5 7beb86b8078fb57c8a0438a2bc6830a1
BLAKE2b-256 ec82b08f9570135e4099a98b7959a22b32ed2191dac26a22be5407bcb9f49a3a

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page