Open-source, self-hosted observability, risk signals, and behavior governance for autonomous AI agents. Guard what they do before they do it.
Three steps from install to full agent observability.
npm install dashclaw # or pip install dashclaw
Zero dependencies. Works with Node.js and Python agents.
const claw = new DashClaw({
apiKey: '...',
agentId: 'my-agent',
})One constructor. Your API key scopes all data.
with claw.track(action='deploy'): # ... actions stream to # dashboard in real-time
Actions, signals, and costs appear instantly via SSE.
Built for teams running autonomous AI agents in production.
Watch your agents think and act live. Server-Sent Events (SSE) stream actions, signals, and costs directly to your mission control dashboard.
Automatic detection of autonomy spikes, repeated failures, stale loops, assumption drift, and more. No configuration required.
Real-time financial tracking. See "Cost per Goal" and "Burn Rate" for every model (GPT-4o, Claude 3.5, etc.) instantly.
Control agent behavior with logic rules (rate limits) or natural language policies ("Don't delete system files").
Zero-dependency clients for both ecosystems. Drop-in compatible with LangChain, CrewAI, and AutoGen.
Structured handoff documents for continuity between agent sessions.
Capture key points and organize context into threads for long-running topics.
Track unresolved dependencies, pending approvals, and blockers across agents.
Log what agents assume, validate or invalidate, and catch drift early.
18 regex patterns detect API keys, tokens, and PII before data leaves your system.
Track memory file counts, duplicates, stale facts, entities, and topics over time.
Aggregated daily summary from actions, decisions, lessons, content, and goals.
Install from npm. Zero dependencies. Works with any Node.js agent framework. Actions, handoffs, context, snippets, messaging, security scanning, and more.
Automatic detection of problematic agent behavior. No configuration required.
Agent taking too many actions without human checkpoints
Critical actions without sufficient review
Same action type failing multiple times
Open loops unresolved past their expected timeline
Assumptions becoming stale or contradicted by outcomes
Assumptions not validated within expected timeframe
Actions stuck in running state for over 4 hours
Team management, audit trails, webhooks, and more — built in from day one.
Invite links, role-based access (admin/member), and workspace isolation.
HMAC-signed webhook delivery plus email alerts via Resend for signal notifications.
Every admin action logged — key creation, invites, role changes, and usage activity.
4-step checklist: create workspace, generate key, install SDK, send first action.
Full org isolation with API key scoping, per-agent settings, and org management.
20+ Python CLI tools for local ops with optional --push sync to the dashboard.
Python CLI tools that run alongside your agent. Local-first with SQLite storage. Add --push to sync anything to your dashboard.
Log decisions, lessons, and outcomes. Track what worked and why.
learner.py log "Used JWT" --push
Key points, threads, and session continuity documents.
context.py capture "Dark theme" --push
Scan memory files, track entities, detect stale facts.
scanner.py scan ~/.agent/memory --push
Goal milestones, contacts, interactions, and follow-ups.
goals.py add "Ship auth" --push
Outbound content filtering, session isolation, audit logging.
outbound_filter.py scan message.txt --push
Reusable code snippets with search, tags, and use tracking.
snippets.py add "retry logic" --push
Install the SDK, send your first action, and see signals on the dashboard. Open-source and self-hosted.