Open source · MIT license · No cloud

Tells you if your AI agents are working, what they're doing, and where your money goes.

20+ features. One install. 60 seconds to first health data. Your data never leaves your machine.

$ pip install observeco && observeco dashboard click to copy

Works with Hermes, OpenClaw, Ollama, LangChain, CrewAI, or any framework.

localhost:8080
ObserveCo dashboard showing fleet health, token breakdown, and error timeline

20+ features. Auto-detects your agent stack.

Everything you need to know if your agents are alive, what's in their context, and when something breaks. Plus webhook ingestion from Slack, Discord, and Telegram.

💓

Pulse Check

Agent liveness — alive, dead, or errored. Every 30 seconds. Zero config.

Free

Circuit Breaker

N-failure trip → auto-block → cooldown. Stops cascade failures before they spread.

Free
🛡️

Safety Guard

99.7% noise reduction. Only surfaces real issues, not flapping or transient errors.

Free
🔧

Heal Button

One-click restart for dead agents. Manual trigger, you're in control.

Free
🧠

Token Breakdown

See exactly what's in your agent's context — identity, skills, memory, tools, guidance.

Free
📈

Drift Tracking

7-day rolling token trend per component. Know before context bloat becomes a problem.

Free
🔍

Skill Audit

Find bloated, duplicate, or unused skills eating your context window.

Free
🌱

Memory Garden

Duplicates, contradictions, stale entries — find and fix memory bloat.

Free
📊

Context Profiler

See what's loading into your agent's context — MEMORY.md, skills, workspace.

Free
⏱️

Error Timeline

Full error history with context snapshots. Not just logs — annotated events.

Free
🔔

In-Dashboard Alerts

See alerts when you open the dashboard. Shows the discovery gap — how late you found out.

Free
💾

Local SQLite

All data on your machine. Zero cloud. Zero telemetry. Your agents' data stays yours.

Free
🌐

Webhook Ingestion

Receive events from Slack, Discord, and Telegram. Translates to OEF → risk engine → alerts.

Free
🔀

Event Pipeline

Tool calls flow through risk classification → session log → alert dispatch → circuit breaker.

Free

Rate Limiting

Built-in 429 handling with exponential backoff. Slack, Discord, Telegram calls never silently fail.

Free
📮

Dead Letter Queue

Failed events are stored, not dropped. Retry, resolve, or purge from the API.

Free
🔐

SAML SSO

Enterprise single sign-on with XML signature validation. Just-in-time user provisioning.

Enterprise
🪪

OAuth2 + Sessions

Google, GitHub login. Sessions persist across restarts via SQLite.

Free
🔒

Encrypted Tokens

API tokens encrypted at rest via OS keychain. No plaintext secrets on disk.

Free
🔄

DB Migrations

Versioned schema migrations. Upgrade safely — no data loss, no breakage.

Free
💓

Watch Self-Check

Heartbeat file every cycle. External monitors can detect if the daemon is stuck.

Free

60 seconds to first health data.

01

Install

pip install observeco — one command, no dependencies to manage.

02

Discover

Auto-detects your agents from config files. Zero configuration if you use Hermes or OpenClaw.

03

Monitor

Dashboard shows fleet health, token breakdown, drift trends, and errors in real time.

$ pip install observeco
✓ Installed observeco-0.1.0
 
$ observeco pulse check
Agent Fleet Health
● hound alive · 30s ago · 12,400 tokens
● kepler alive · 30s ago · 8,200 tokens
● aleph alive · 30s ago · 15,600 tokens
● dreamer degraded · 2 missed heartbeats
● raven dead · no response in 5m
 
$ observeco dashboard
Dashboard running at http://localhost:8080

Built for agent operators, not infrastructure teams.

ObserveCo
Alternatives
Setup time
60 seconds — pip install, done
2+ hours — Grafana + Prometheus
Understands tokens
Per-component breakdown, drift, cost
Generic metrics, no context concept
Memory hygiene
Duplicates, contradictions, debt score
Not available
Circuit breakers
Built-in, auto-cooldown
Manual configuration
Webhook ingestion
Slack, Discord, Telegram → OEF → risk engine
Manual integration required
Rate limiting
Built-in 429 retry + backoff
Silent failures on rate limits
Dead letter queue
Failed events stored, retryable
Events silently dropped
SSO / SAML
Built-in with signature validation
Not available
Cost
Free forever — MIT license
$15-59/mo — per host
Data
Local — SQLite, no cloud
Cloud — your data leaves your machine

v0.1 is just the beginning.

We ship fast. Every week, new capabilities. The dashboard that sees problems also fixes them.

v0.2 · Now

20+ features — monitoring + integration

Pulse check, circuit breakers, token breakdown, drift tracking, memory garden, skill audit, heal button, error timeline, webhook ingestion, event pipeline, rate limiting, dead letter queue, SAML SSO, OAuth2, encrypted tokens, DB migrations.

v0.3 · Next

Auto-heal + push alerts

Watch daemon auto-restarts dead agents. Telegram, webhook, and email alerts when agents break. Close the discovery gap.

v1.0 · Soon

OpenClaw runtime plugin

40-60% fewer tokens per turn. Intent-aware context loading that only loads what your agent needs, when it needs it.

See your agents in 60 seconds.

No signup. No account. Just open source.

$ pip install observeco && observeco dashboard

GitHub · MIT License · Built for the AI agent community