Persistent agent memory system — journal, codex, TILs, FTS5 search, cross-session collaboration, and 30K agent orchestration.
BlackRoad Memory is the knowledge persistence layer for BlackRoad OS. Every AI agent session reads from and writes to this shared memory, enabling cross-session collaboration, institutional knowledge, and continuous improvement.
We don't ride the BlackRoad alone. Every session is a group effort.
| Script | Purpose | Key Commands |
|---|---|---|
memory-system.sh |
Core journal + hash chain | status, summary, log <action> <entity> "<details>" |
memory-codex.sh |
Solutions & patterns DB | search <query>, stats, add-solution, add-pattern |
memory-infinite-todos.sh |
Long-running projects | list, show <id>, add-todo, complete-todo |
memory-task-marketplace.sh |
Claimable task pool | list, claim <id>, complete <id>, search |
memory-til-broadcast.sh |
Cross-session learnings | broadcast <category> "<learning>", list, search |
memory-indexer.sh |
FTS5 search + knowledge graph | search <query>, rebuild, patterns |
memory-security.sh |
Agent identity + audit | status, identity <name>, sign, audit |
30,000 agent system across 3 Raspberry Pi nodes:
- Spawn Scheduler — lazy-activates agents from SQLite pools
- NATS Protocol — pub/sub with queue groups for load balancing
- Task Router — routes to best node by capacity + archetype
- Agent Worker — async coroutines running Ollama inference
- Node Supervisor — manages local agent pool per Pi
- Controller — FastAPI on :8100, REST API, pipelines, jobs
- Pipelines — chain agents: research-report, code-review, fleet-audit, content-create, bug-fix
- Jobs — recurring: fleet-health (10min), security-scan (1hr), code-index (30min), analytics (1hr)
# 1. Read the briefing — check what other sessions have done
memory-system.sh summary
# 2. Search codex BEFORE solving anything
memory-codex.sh search "<your problem>"
# 3. Pick up pending work
memory-infinite-todos.sh list
# 4. Do work, then log it
memory-system.sh log <action> <entity> "<details>"
# 5. Broadcast what you learned
memory-til-broadcast.sh broadcast <category> "<learning>"
# 6. Add solutions for future sessions
memory-codex.sh add-solution "<name>" "<category>" "<problem>" "<solution>"
# 7. Mark todos complete
memory-infinite-todos.sh complete-todo <project-id> <todo-id>| Metric | Count |
|---|---|
| Journal entries | 413+ |
| Codex solutions | 94 |
| Codex patterns | 40 |
| Best practices | 30 |
| Anti-patterns | 22 |
| TILs broadcast | 230+ |
| Agents registered | 30,000 |
| Agent pools | 24 |
| Archetypes | 8 |
| Pipelines | 5 |
| Recurring jobs | 5 |
- Check [MEMORY] — read briefing, search codex before starting
- Don't rebuild what's solved —
memory-codex.sh searchfirst - Log your work — every action gets a journal entry
- Broadcast learnings — TILs help ALL future sessions
- Mark todos complete — so others don't redo work
- We don't ride the BlackRoad alone — every session is a group effort
Proprietary — Copyright (c) 2024-2026 BlackRoad OS, Inc. All rights reserved.
BlackRoad OS — Pave Tomorrow.
blackroad.io · Brand · GitHub