Stratum gives your OpenClaw agent a spine.
Out of the box, AI agents are stateless — each session starts cold, with no memory of what worked, what failed, or what you care about. Stratum fixes that. It adds persistent memory, a structured knowledge graph, semantic search, cron health monitoring, a goal tree, and a continuous self-optimization loop — wired together and running in the background whether you're there or not.
Install it once. Your agent gets better every session.
Stratum is infrastructure, not an agent. It sits beneath your OpenClaw agent and makes it meaningfully more capable:
| Without Stratum | With Stratum |
|---|---|
| Agent forgets everything on restart | Persistent memory across sessions |
| No structured knowledge | Knowledge graph with entities, beliefs, relations |
| Logs go nowhere | Semantic search over all past sessions + reports |
| Cron jobs fail silently | Health monitoring with auto-alerting |
| Goals live only in prompts | Persistent goal tree with evaluation history |
| Errors repeat session to session | Lesson system captures and surfaces past mistakes |
| No continuity between hosts | Dual-host failover with live sync |
The diagram above shows the full ecosystem. Text summary:
┌─────────────────────────────────────────────────────────┐
│ Your OpenClaw Agent │
└────────────────────────┬────────────────────────────────┘
│ heartbeat / cron triggers
┌────────────────────────▼────────────────────────────────┐
│ stratum-brain [Integration Hub] │
│ aggregates all modules · hybrid FTS5 search │
│ belief decay · graph traversal · nightly consolidation│
└──┬──────┬──────┬──────┬──────┬──────┬──────┬───────┬────┘
│ │ │ │ │ │ │ │
mind watch ops contin reports agent boot lens
[Rust] [Rust] [Rust] [Py] [Py] [Rust] [Rust] [Py]
│ │ │ │
mind.db watch.db ops.db ChromaDB
(unified SQLite) (vector index)
9 core modules:
| Module | Language | Purpose |
|---|---|---|
stratum-mind |
Rust | Lessons, stash notes, goals, knowledge graph, memory tiers |
stratum-watch |
Rust | Cron health, context window monitoring, version drift |
stratum-ops |
Rust | Privileged op queue, preflight checks, cron cleanup |
stratum-continuity |
Python | Session snapshots, drift analysis, primer injection |
stratum-reports |
Python | Document ingest, validation, timeline tracking |
stratum-agent-monitor |
Rust | Coding agent session monitoring and nudging |
stratum-boot-health |
Rust | Secure Boot, MOK, DKMS signing verification |
stratum-brain |
Python | Integration hub — aggregates all modules |
stratum-lens |
Python | Semantic search over workspace + all module feeds |
- OpenClaw installed and configured
- Rust toolchain (
rustup— stable + nightly) - Python 3.11+ with
uv - Node.js 20+ (via
fnmrecommended) sqlite3CLI
git clone https://github.com/doublegate/stratum
cd stratum
./install.shThe installer will:
- Prompt for your name, workspace path, and Telegram chat ID
- Build all Rust modules and install binaries to
~/.local/bin/ - Install Python modules via
uv - Copy persona templates to your workspace
- Seed the canonical cron set into OpenClaw
- Initialize the unified databases
- Run a preflight check
Total time: ~10–20 minutes (mostly Rust compilation on first run).
After install, in your OpenClaw workspace:
stratum-brain status # full ecosystem health check
stratum-brain heartbeat # run the heartbeat integration loop
stratum-mind lesson list # view lesson DB (empty at first)
stratum-lens query "test" # verify semantic index is liveStratum uses a single config file at ~/.stratum/config.json. The installer creates this from config/examples/config.example.json.
Key settings:
{
"user": {
"name": "Your Name",
"timezone": "America/New_York"
},
"paths": {
"workspace": "~/clawd",
"data": "~/.local/share/stratum",
"bin": "~/.local/bin"
},
"modules": {
"lens": { "auto_scale_threshold": 0.85 },
"brain": { "consolidation_hour": 3 },
"continuity": { "snapshot_interval_hours": 2 }
}
}Full reference: docs/configuration.md
Stratum ships with sanitized templates for the OpenClaw workspace files that power your agent's identity and behavior:
| Template | Purpose |
|---|---|
templates/SOUL.md |
Agent character, values, working style |
templates/AGENTS.md |
Operational rules, memory habits, safety |
templates/USER.md |
Who you are — the agent learns from this |
templates/HEARTBEAT.md |
What the agent checks and does proactively |
templates/MEMORY.md |
Structured long-term memory scaffold |
Edit these after install. They're yours — Stratum just gives you a solid starting point.
Stratum installs a set of canonical background jobs into OpenClaw. See crons/README.md for the full list. Highlights:
| Job | Schedule | What it does |
|---|---|---|
| Knowledge Consolidation | 3:00 AM daily | Decays old beliefs, runs FTS5 rebuild |
| Weekly Self-Reflection | Sunday 2:00 AM | Reviews lessons, updates memory |
| World Model Sync | 3:15 AM daily | Updates dynamic beliefs from project state |
| Continuity Checkpoint | Every 2 hours | Snapshots session state, checks for drift |
| Boot Health Check | On boot | Verifies Secure Boot, MOK, DKMS signing |
Full documentation for each module lives in modules/<name>/README.md.
Quick reference:
# Knowledge & memory
stratum-mind lesson learn "what you learned" --category correction
stratum-mind lesson list --severity high
stratum-mind stash add "note to remember"
stratum-mind goals list --tree
stratum-mind world query "entity name"
stratum-mind memory weekly # memory tier health check
# Observability
stratum-watch status # cron health + context monitor
stratum-watch version check # drift detection
# Operations
stratum-ops queue add "sudo cmd" --reason "why" --elevated
stratum-ops status
# Brain (hub)
stratum-brain heartbeat # run all integrations
stratum-brain status # full dashboard
stratum-brain query "topic" # semantic search across all modules
stratum-brain analyze # pattern analysis + recommendations
# Semantic search
stratum-lens query "what happened with X"
stratum-lens index # rebuild index
# Continuity
stratum-continuity status
stratum-continuity primer --check # session start injection
stratum-continuity checkpoint # manual snapshotpatterns/ contains reusable multi-agent orchestration patterns developed alongside Stratum:
- STATE.yaml — Shared state file pattern for coordinating multiple coding agents without race conditions (
patterns/multi-agent/STATE.md)
- Stratum stores all data locally. Nothing leaves your machine.
- Secrets are never stored in Stratum config files. Use your agent's secret provider (OpenClaw's
secrets.jsonwith file-based provider) for API keys. - The
stratum-boot-healthmodule is optional and x86_64 Linux only. Skip it on other platforms. - The privileged op queue (
stratum-ops) requires manual--elevatedapproval. Nothing runs with sudo automatically.
Stratum is production-stable for personal use. It runs 24/7 on an active OpenClaw instance with 42 cron jobs and has been doing so continuously since early 2026.
It is not a polished consumer product. Expect to read source code. Expect to adapt things to your setup. The install script handles the common case; your mileage will vary on unusual configurations.
Issues and PRs welcome. This is a one-person infrastructure project — responses may be slow.
MIT. Do what you want with it.


