ObserveCo fleet
Tells you if your AI agents are working, what they're doing, and where your money goes.
🔓 FREE
MIT License Free forever · No cloud · Apache 2.0 compatible
🔒 Local only — this dashboard is accessible to any process on this machine. Do not expose to the internet.
👋 First time here? Detect agents or add an agent to start monitoring your fleet.

🧠 Brain Analysis

See what feeds your agents. See what you can save.
13 agents FREE

Token Breakdown per component

0 total
What each part means
skills = task instructions
tools = functions and APIs
memory = user context and history
guidance = behavioural rules
identity = role and personality

What Token Optimiser Can Save You

50 turns/day avg
🔒 Full compression saves more — that's extra per year
Pro gives you Full compression + Auto-Watch + Token Optimiser + never-pruned history

Drift & Usage 7-day trend

Component drift
Per-turn token usage last 24h FREE
24h agoNow

Compression see the diff, then apply

Lite: compress guidance blocks · Full: +memory culling +skill dedup +context refactor
Before compression
0 tok
Select an agent and run Preview to see the diff.
After (Lite compression)
0 tok —
 
Step 1: Preview
See what changes before applying. No file is modified.
Step 2: Apply
Write compressed version to agent's SOUL.md. Backup created automatically.

🧠 Token Optimiser PRO learns from 200+ turns

Beyond simple compression. The Optimiser analyses every turn your agents take — which skills are used, which rules trigger, which memory gets referenced — then surgically removes what's unused and restructures what remains.

🔒

📊 Learning progress

58% — learned from 116 turns Goal: 200 turns
Skills never triggered3 of 8
Guidance rules stale2 of 5
Recent compressions—
At 200 turns, Optimiser will recommend what to prune — projected 8-12% additional savings on top of compression.

📈 Projected savings

-22%
Lite (current)
-35%
Full (available)
-43% to -47%
Projected with Optimiser after 200 turns
Lite compression + Optimiser pruning = deeper savings
→ At 50 turns/day, you'll have enough data in ~2 days
🔒 Unlock Token Optimiser + Auto-Watch + Full compression with Pro →
🔓 FREE
Optimiser Lite
• Compress guidance blocks
• Per-agent breakdown & drift
• 24h per-turn timeline
• 7-day component trends
22% savings per turn
🔒 PRO
Optimiser Full
• Full compression (memory + skills + context)
• Auto-Watch daemon
• Token Optimiser (learns from 200 turns)
• Never-pruned history & fleet comparison
Upgrade → Up to 47% savings

🛠️ Auto-Heal

See how self-healing can recover agents automatically.
🔓 Free (Manual)
10:32 Agent down
10:44 ⚠️ You notice via dashboard
10:58 💻 You investigate manually
11:19 ❌ Agent still down — 47 min
...
12:30 ✅ You restart — back in 1h 58m
Outcome: 1h 58m downtime
$0 saved You did all the work. Downtime costs stack while you investigate.
🛡️ Pro — L1 Auto-Heal
10:32 Agent down
10:32 ⚡ Detected instantly
10:32 🔄 Auto-restart triggered
10:33 ✅ Agent back online — ~1 min
Outcome: ~1 min downtime
$0.02 saved / event Instant detection, auto-restart. You don't lift a finger.
🧠 Pro — L2 Proactive
10:31:45 📊 Memory pressure rising
10:31:50 🔧 Preemptive rotation
10:32:00 ✅ Crisis averted — no downtime
Outcome: 0 downtime
$0.03 saved / event Predicted failure before it happened. Zero impact.
Free Pro L1 Auto-Heal Pro L2 Proactive
Detection speed Manual — you notice Instant — pulse detects Predictive — before failure
Recovery action You SSH & restart Auto-restart process Auto-rotate + notify you
Avg downtime ~2 hours ~1 minute 0 — prevented
Cost / agent / month $0 (included) $0 (included in Pro) $0 (included in Pro)
Circuit breaker ❌ Not available ✅ Stops cascade failures ✅ + predictive alerts
Heal Log recent events
🔒 Auto-Heal is Pro
Free: manual diagnose only. Pro auto-detects, auto-restarts, and predicts failures before they happen.
🧠 L2 Proactive Monitoring trend-based pre-emptive detection
📊 No trends recorded yet. Click "Run L2 Scan" to start.

🔔 Push Alerts

Know when something breaks — before your users tell you.
🔓 Free (Dashboard Only)
🔴Agent DownCritical
hound has not checked in for 14 minutes
Found when you open dashboard
You discover it later
~2h delayed No notification — you only see it when you check the dashboard.
🛡️ Pro — Silent Auto-Heal
🟢Agent Down → HealedAuto
hound auto-restarted in 47s
No notification needed — you're never woken up
You never know (that's good)
Auto-healed Agent recovered before you'd even notice. Zero alerts to your phone.
📲 Pro — Alert on Exhaustion
🟡Circuit Breaker TrippedWarning
dreamer: circuit tripped after 5 failures — auto-retry in 4h
📱 Pushed to Telegram · 2s latency
🔍 Discovery gap: ~10 min → you know in real time
Know instantly
2s latency Push notification to Telegram/Slack/email. You decide if it needs attention.
Free Pro Silent Pro Alert
Detection Dashboard only Auto-heal silently Push + Auto-heal
Delivery ❌ None ✅ None needed ✅ Telegram / Slack / Email
Discovery gap ~2h 0 (auto-healed) ~2s
Cost $0 $0 (Pro) $0 (Pro)

Channel Configuration FREE

🔔
In-Dashboard
Alerts appear in the dashboard alert panel
Active
📱
Telegram
Push notifications to your Telegram channel
Pro
💬
Slack
Post alerts to a Slack channel
Pro
📧
Email
Email notifications for critical alerts
Pro
💡
Discovery gap matters. Every minute between an agent going down and you finding out costs time, money, and user trust. Pro reduces that gap from hours to seconds — and Silent mode eliminates it entirely.
🔒 Push Alerts are Pro
Free: dashboard alerts only. Pro adds Telegram/Slack/email push — and Silent mode that heals before you need to know.
📡 Subscriptions active alert channels
📋 Delivery Log recent alert deliveries

🔌 ClawForge Plugin

Intent-aware context loading — save 40-60% per turn inside OpenClaw agents.
0
Turns Tracked
0
Tokens Saved
0%
Avg Reduction
0
Sources Loaded
Recent Hook Events from OpenClaw agents

Install @observeco/clawforge-plugin

# Install the plugin
npm install @observeco/clawforge-plugin

# Add to OpenClaw config
{
  "plugins": ["@observeco/clawforge-plugin"]
}

# Activate (no restart needed)
observeco plugin log ... (or POST /api/plugin-log)