CharmOS

August 14, 2026 · View on GitHub

The first open-source framework that manages both your human contacts AND your AI agents as first-class relationships, on top of a personal "second brain", driven by one decision loop: RDR.

C(H+A)RM = CRM for Humans + Agents Relationship Management. RDR = Recall to Deep Research to synthesis. The loop that turns memory into decisions.

What this is

Three things most people keep in separate silos, unified into one open framework:

  1. Second Brain to a private, local-first knowledge base (markdown vault) with semantic recall (embeddings + reranker), an always-on per-turn memory ledger, and a graph/associative recall layer.
  2. C(H+A)RM to a relationship layer that treats people and AI agents alike as contacts you maintain: notes, history, warm intros, follow-ups.
  3. RDR to the method that ties them together: Recall what you already know, run Deep Research to fill the gaps, then synthesize into a Decision Memo before you act.

Why it is different

The ecosystem is rich but siloed:

SpaceExamplesWhat they miss
Personal knowledge (PKM)AFFiNE, Logseq, Khoj, Quivrno agent memory, no contacts
Agent / LLM memoryMem0, Letta, Cognee, Graphitino personal vault, no human CRM
Personal CRMMonica, Twentyignore AI agents entirely

No tool spans personal notes to agent memory to human CRM, and none prescribe a structured decision loop. That intersection, plus RDR, is the whole point.

Who it is for

Founders, researchers, and operators who want a durable "digital twin" of how they think and who they know, that they fully own, can repair themselves, and can grow over years.

Architecture (high level)

        ┌──────────────────────────────────────────────┐
        │                   RDR LOOP                     │
        │  Recall  ->  Deep Research  ->  Synthesis      │
        └───────▲───────────────────────────────┬───────┘
                │                                 │
        ┌───────┴────────┐               ┌────────▼────────┐
        │  SECOND BRAIN  │               │    C(H+A)RM     │
        │ vault + RAG +  │◄─────────────►│ humans + agents │
        │ memory ledger  │   shared      │ as contacts     │
        │ + graph recall │   knowledge   │ intros, history │
        └───────▲────────┘               └────────▲────────┘
                │                                 │
        ┌───────┴─────────────────────────────────┴───────┐
        │   Import pipelines (chat / mail / calls / docs)   │
        │   Skills (composable agent commands)             │
        └──────────────────────────────────────────────────┘

Status

Manifest-first, now with runnable modules. This repository shipped the vision, architecture, and docs first (see MANIFESTO.md and /docs); real code is now landing under modules/:

  • modules/turnstate/ (v0.2.0) — the always-on memory ledger: per-turn deterministic working state, hook + self-healing backfill.
  • modules/rdr/ (v0.3.0) — the RDR loop as a CLI: recall → research → memo, built on the TurnState ledger.
  • modules/eval-harness/ (v0.1.0) — grade your agent fleet against rules it must never break. Turns a multi-agent event log into reproducible traces and scores them against explicit safety invariants. Includes our own production results: human-gate before a risky commit 100%, independent verify before commit 7.7% — we publish both.

All deterministic, zero-LLM-token, pure stdlib. More modules follow.

Roadmap

Now — v0.4.0. The three modules above, each versioned on its own, plus the manifesto and docs that came first.

Next, from CHANGELOG.md: a Second Brain reference implementation (vault + RAG + reranker), the graph/associative recall layer with entity-vs-theme gating, the C(H+A)RM relationship layer (humans and agents as contacts), and the import pipelines.

We version with SemVer, and every noticeable change ships as a new release — a shipped module is a release, cut when the work lands rather than when someone remembers. The release feed is how you tell what is real here from what is still a manifesto.

Privacy (read this first)

This framework operates on deeply personal data. No real personal data is included in this repository. Everything under /examples is synthetic. If you self-host, your data stays yours and local. See docs/privacy.md.

Cite this work

If this repo shows up in your research, cite it via CITATION.cff (GitHub's "Cite this repository" button). Author: Anton Dziatkovskii (ORCID 0000-0001-7408-3054, GitHub @tonydzi) — one spelling everywhere, in publications, commits and CITATION.cff.

AI contributors

This project is built by a human + AI team, and the git log says so: Claude writes most of the code, Codex and Grok review it, Gemini feeds the research. Each is credited on a commit only if its output changed that commit's content — no decorative credits. Lab-wide policy, one source for every repo: AI-CONTRIBUTORS.md.

License

Apache License 2.0. Permissive, with an explicit patent grant. An open-core path (optional paid layer/hosting) may follow, but the core stays open.

Built by Palo Alto AI Research Lab. Contributions and discussion welcome once V1 docs settle.

About & contact

Questions, war stories, or you want to run this on your own fleet:

🧩 One piece of a working system

This repository is one piece lifted out of a live operation: one non-technical founder, an AI cofounder, and a fleet of machines that reach consensus with each other and wake the human only for money or the irreversible. It was extracted after it survived production, not written as a demo — and it runs on its own: nothing here phones home to the rest.

See how the whole thing fits together → SYSTEM.md

Its closest neighbours in the governance layer: claude-bible · agent-leash