observational-memory 0.6.1

Cross-agent observational memory and local search for Claude Code, Codex CLI, Cowork, and Hermes Agent

Observational Memory

Local memory for the agents you already use.

Observational Memory, or om, gives Claude Code, Codex, Claude Cowork, and Hermes one shared memory on your machine. It watches agent transcripts, writes useful notes into local Markdown files, and gives new sessions a compact startup context. You can search that memory later, export reviewed memory bundles for hosted platforms, or opt in to encrypted multi-machine sync with OM Cluster.

The current release is v0.6.1. It includes:

• budgeted startup context through om context
• first-class recall through om recall
• richer reflection metadata and host-memory controls
• OM Cluster relay operations and health checks
• public-safe cluster validation docs
• Windows, macOS, and Linux install paths

Quick Install

macOS with Homebrew:

brew install intertwine/tap/observational-memory
om install
om doctor

Linux, macOS, or Windows with uv:

uv tool install observational-memory
om install
om doctor

Install the optional enterprise auth extras if you use Anthropic through Vertex AI or Bedrock:

uv tool install "observational-memory[enterprise]"

What It Does

om keeps four main memory files under your local data directory:

File	Purpose
observations.md	Recent notes from sessions and checkpoints.
reflections.md	Longer-term facts, preferences, decisions, and active work.
profile.md	Compact stable context for startup.
active.md	Compact current context for startup.

Those files are plain Markdown. You can read them, back them up, and search them.

Default paths:

Platform	Memory directory	Config directory
macOS / Linux	~/.local/share/observational-memory/	~/.config/observational-memory/
Windows	%LOCALAPPDATA%\observational-memory\	%APPDATA%\observational-memory\

How Memory Flows

flowchart LR
    A["Claude Code, Codex, Cowork, Hermes logs"] --> B["om observe"]
    C["Claude auto-memory files"] --> D["search index"]
    B --> E["observations.md"]
    E --> F["om reflect"]
    D --> F
    F --> G["reflections.md"]
    G --> H["profile.md + active.md"]
    H --> I["om context startup pack"]
    G --> J["om recall / om search"]

First Week Workflow

1. Install om.
2. Run om install and answer the provider questions.
3. Run om doctor.
4. Start using Claude Code or Codex normally.
5. Search memory when you need it:

om recall --query "current project status"
om search "release checklist"

6. Check generated startup context:

om context --for codex --cwd "$PWD" --task "finish docs"

Guides

Start here:

• Documentation index
• Install and setup
• Platform integrations
• Search, recall, and startup context
• Configuration
• OM Cluster sync
• OM Cluster validation checklist
• Host memory coexistence
• Maintainer guide

Agent Support

Host	Current support
Claude Code	Hooks for startup context and checkpoints.
Codex	Hooks-first startup and Stop checkpoints, with an AGENTS fallback.
Claude Cowork	Local plugin on macOS with hooks and /recall.
Hermes	Manual session-log ingestion. A first-class Hermes plugin is planned in plans/hermes-first-class-plugin.md.
ChatGPT / Claude Managed Agents	Reviewed export bundles through om export; om does not silently write hosted memory.

Common Commands

om status
om doctor
om observe --source codex
om reflect
om recall --query "what was decided about sync?"
om recall --handle startup:active
om search "preferences" --json
om export --target chatgpt
om export --target claude-managed-agents --output ./om-claude-memory

OM Cluster is off until you initialize or join a cluster:

om cluster init --name "Personal Memory" --transport filesystem:~/Sync/om-cluster --import-existing
om cluster invite --expires 10m
om cluster join "omc1:..."
om cluster requests
om cluster approve join_...
om cluster sync
om cluster status

Do not sync ~/.local/share/observational-memory/ directly with Dropbox, iCloud, Syncthing, rsync, or a NAS. Use the cluster transport directory instead.

Architecture At A Glance

The short version:

• om observe turns transcripts into recent notes.
• om reflect turns recent notes into durable memory.
• om context gives agents a bounded startup pack.
• om recall and om search retrieve more when the startup pack is not enough.
• om export prepares reviewed memory seed bundles for hosted systems.
• om cluster syncs encrypted records across machines when you opt in.

Release State

v0.6.1 is the current release. It hardens the 0.6.0 OM Cluster release, adds budgeted startup context and first-class recall, and updates the user docs for day-to-day setup.

Before the next release, maintainers should run:

make check
uv run ruff check .
uv run ruff format --check .
uv run pytest

See docs/MAINTAINERS.md for the full release workflow.