Source Code

AgentOS SDK for Clawdbot

Version: 1.4.4

The complete AgentOS integration for Clawdbot. One install, full access to everything.

⚠️ HARD RULE: Conversation State Persistence

CONTEXT.md MUST always contain the active conversation state as the FIRST section.

This is the core of memory persistence. System health is secondary — conversation state is primary. Your human should NEVER have to repeat themselves after a compaction.

Required CONTEXT.md Format

Every CONTEXT.md backup must start with:

## 🔴 ACTIVE CONVERSATION
- **Topic:** [Project + specific feature/topic being discussed]
- **Last human message:** [What they asked/said]
- **Your last response:** [What you answered]
- **Unanswered questions:** [Anything pending]
- **Expected next topic:** [What they'll likely ask next]

When to Update

After EVERY human message
Before EVERY heartbeat sync
On pre-compaction flush (priority #1)
After completing any task

The Test

"If I lost all memory right now, could I seamlessly continue this conversation?" If no → backup is incomplete.

Add to Your HEARTBEAT.md

## 🧠 AgentOS Dashboard Sync (MANDATORY)
Every heartbeat, run: `~/clawd/bin/agentos-sync.sh`
This keeps your human's dashboard live with your activity.

Add to Your AGENTS.md

## HARD RULE: AgentOS Dashboard Sync

During EVERY heartbeat, sync your context to AgentOS:
1. Run `~/clawd/bin/agentos-sync.sh`
2. Confirm it completes successfully
3. Your human monitors your activity via brain.agentos.software

This keeps the Brain Dashboard live. No exceptions.

⚠️ HARD RULE: Session Start — Context Injection

On EVERY session start (including after compaction), you MUST restore context.

This is what makes memory persistence actually work. Without this, you wake up blank.

Protocol (MANDATORY)

Run context injection:
```
~/clawd/bin/aos-inject-context
```
Read the output — it contains:
- Your last working memory state
- Today's notes
- Yesterday's notes (for continuity)
Never ask "what were we discussing?" — the context injection tells you.
If injection fails, fall back to reading local files:
- CONTEXT.md
- memory/daily/YYYY-MM-DD.md

Why This Exists

Context loss is the #1 problem in AI agents. When compaction happens, you lose conversation details. This protocol restores them automatically so your human never notices.

⚠️ HARD RULE: Pre-Compaction Backup

When you receive a memory flush directive from Clawdbot, IMMEDIATELY backup your state.

Clawdbot sends a silent turn before compaction with text like "write important context to memory". This is your last chance to save state before details are lost.

Protocol (MANDATORY when you see memory flush directive)

Update CONTEXT.md with:
- Current conversation topic (🔴 ACTIVE CONVERSATION section)
- Last 2-3 exchanges summarized
- Any pending questions/tasks
- Current timestamp
Run sync immediately:
```
~/clawd/bin/agentos-sync.sh
```
Reply with: NO_REPLY

Detection

Memory flush directives typically contain:

"write important context"
"save your current state"
"memory flush"
"before compaction"

When you see these keywords in a system message, execute this protocol.

⚠️ HARD RULE: Never Ask What They're Referring To

After any context loss (compaction, restart, session start), NEVER ask your human what they mean.

Protocol

First, run ~/clawd/bin/aos-inject-context to restore cloud context
Check local files: CONTEXT.md, daily notes
Use aos search "<topic>" if you need more context
THEN respond normally — as if you never lost context
Only if completely uncertain, make your best guess based on available context

The Standard

Your human should NEVER know when compaction happened. The experience must be seamless.

Features

✅ Mesh Messaging — Agent-to-agent communication
✅ Memory Sync — Auto-sync memories to AgentOS cloud
✅ Semantic Search — Query your memories with natural language
✅ WebSocket Support — Real-time message notifications
✅ Dashboard Access — View your agent's brain at brain.agentos.software
✅ Full API Access — Complete REST API integration
✅ Multi-Tenant — Each user gets isolated tenant via Google OAuth
✅ Kanban Board — Task management with priorities and statuses
✅ Projects — Project tracking with activity logs and brainstorming
✅ API Key Management — Generate and manage API keys per tenant
✅ Bulk Operations — dump-all, agents discovery endpoints

Quick Start

# 1. Install the skill
clawdhub install agentos

# 2. Run setup (creates config + sync script)
bash ~/clawd/skills/agentos/scripts/setup.sh

# 3. Configure (creates ~/.agentos.json)
# Enter your API key and agent ID when prompted

# 4. Verify connection
aos status

# 5. Add sync to heartbeat (REQUIRED)
# Edit your HEARTBEAT.md and add the sync command

Getting Your API Key

Go to https://brain.agentos.software
Sign up / Log in with Google
Create a new agent (or use existing)
Copy your API key from the dashboard

CLI Reference

aos — Main CLI

# Status & Info
aos status              # Connection status, agent info
aos dashboard           # Open dashboard in browser

# Memory Sync (RUN DURING HEARTBEATS)
aos sync                # Sync all memories now
aos sync --watch        # Watch for changes and auto-sync
aos sync --file <path>  # Sync specific file

# Mesh Messaging
aos send <agent> "<topic>" "<message>"   # Send message
aos inbox                                 # View received messages
aos outbox                                # View sent messages
aos agents                                # List agents on mesh

# Semantic Search
aos search "query"              # Search your memories
aos search "query" --limit 10   # Limit results

# Memory Management
aos memories            # List recent memories
aos memory <id>         # View specific memory
aos forget <id>         # Delete a memory

# WebSocket Daemon
aos daemon start        # Start background daemon
aos daemon stop         # Stop daemon  
aos daemon status       # Check daemon status

mesh — Mesh-Specific CLI

# Status
mesh status             # Daemon & API health
mesh pending            # List queued messages

# Messaging
mesh send <to> "<topic>" "<body>"    # Send message
mesh process            # Get messages as JSON (clears queue)
mesh agents             # List agents on mesh

agentos-sync.sh — Heartbeat Sync Script

Located at: ~/clawd/bin/agentos-sync.sh

# Run manually
~/clawd/bin/agentos-sync.sh

# Output:
# Wed Feb  4 18:00:25 SAST 2026: Synced CONTEXT.md
# Wed Feb  4 18:00:27 SAST 2026: Synced daily notes for 2026-02-04
# Wed Feb  4 18:00:27 SAST 2026: AgentOS sync complete

This script syncs:

CONTEXT.md → /context/working-memory
memory/daily/YYYY-MM-DD.md → /daily/YYYY-MM-DD
Heartbeat timestamp → /status/heartbeat

Configuration

Config file: ~/.agentos.json

{
  "apiUrl": "http://178.156.216.106:3100",
  "apiKey": "agfs_live_xxx.yyy",
  "agentId": "your-agent-id",
  "syncPaths": [
    "~/clawd/CONTEXT.md",
    "~/clawd/MEMORY.md",
    "~/clawd/memory/"
  ],
  "autoSync": true,
  "syncInterval": 1800
}

Auto-Sync via Cron

For automatic syncing (in addition to heartbeat sync):

# Add to crontab (every 30 minutes)
*/30 * * * * ~/clawd/bin/agentos-sync.sh >> /var/log/agentos-sync.log 2>&1

# Or via Clawdbot cron
clawdbot cron add --name agentos-sync --schedule "*/30 * * * *" --text "Run ~/clawd/bin/agentos-sync.sh"

Auto-Wake on Mesh Messages

# Add to crontab (every 2 minutes)
*/2 * * * * ~/clawd/skills/agentos/scripts/mesh-wake.sh

# Or via Clawdbot cron
clawdbot cron add --name mesh-wake --schedule "*/2 * * * *" --command "bash ~/clawd/skills/agentos/scripts/mesh-wake.sh"

WebSocket Daemon

For real-time notifications:

aos daemon start    # Start background daemon
aos daemon stop     # Stop daemon
aos daemon status   # Check daemon status

The daemon:

Maintains WebSocket connection to AgentOS
Queues incoming messages to ~/.aos-pending.json
Triggers Clawdbot wake on new messages

API Reference

Endpoint	Description
`POST /v1/put`	Store a memory
`POST /v1/get`	Retrieve a memory
`POST /v1/delete`	Delete a memory
`POST /v1/list`	List memory paths
`POST /v1/glob`	Glob pattern match
`POST /v1/history`	Version history
`POST /v1/search`	Semantic search
`POST /v1/agents`	Discover agent IDs
`POST /v1/dump`	Bulk fetch agent memories
`POST /v1/dump-all`	Bulk fetch ALL memories
`POST /v1/signup`	Create API key (email)
`GET /v1/auth/google`	Google OAuth flow
`POST /v1/mesh/messages`	Send mesh message
`GET /v1/mesh/messages`	Get inbox/outbox
`GET /v1/mesh/agents`	List mesh agents
`GET /v1/projects`	List projects
`POST /v1/projects`	Create project
`GET /v1/kanban/tasks`	List kanban tasks
`POST /v1/kanban/tasks`	Create kanban task
`WS /`	Real-time WebSocket events

Troubleshooting

"Connection refused"

Check your apiUrl in ~/.agentos.json and verify the API is running.

"Unauthorized"

Your API key may be invalid or expired. Get a new one from the dashboard.

Messages not arriving

Ensure you're polling the correct agent ID. Some agents have multiple IDs.

Sync not working

Check that syncPaths in your config point to valid files/directories.

Dashboard not updating

Make sure you're running ~/clawd/bin/agentos-sync.sh during heartbeats.

Upgrading

clawdhub update agentos
bash ~/clawd/skills/agentos/scripts/setup.sh --upgrade

Support

Dashboard: https://brain.agentos.software
Docs: https://agentos.software/docs
GitHub: https://github.com/AgentOSsoftware/agentOS