Skip to content

Open-source agent observability

Understand agents.
Improve how humans use AI.

cot is the open-source observability layer for AI agents. It records what your agents do and shows you where to improve security, cost and usability, from real sessions.

View on GitHubBook a call
127.0.0.1:31337/sessions/sess_4c21e8
  1. listening on :31337

INSIGHTS

Detections are checking this session…

events
0
source
claude
  • Claude Code
  • Cursor
  • Codex
  • OpenCode
  • Self-hosted by default
  • Your data stays yours
  • Open source, AGPL-3.0

built in the open

Useful before you ever talk to us.

cot is open source and self-hosted. Install it, connect your agents, and you get traces, detections, search and analysis on your own data.

terminal

$

    observability is the starting point

    You can't improve what you can't see.

    Agents now build, investigate, write and operate alongside us. Once one starts working, it gets hard to tell what actually happened.

    cot gives every agent the same observability layer: sessions, tool calls, files, models, errors and cost, in one place.

    1. 01 ACTIONS
    2. 02 TIMELINE
    3. 03 METRICS
    4. 04 INSIGHTS
    5. 05 TEAMS

    01 · AGENT ACTIONS

    Every dot is one thing an agent did.

    A file read, a shell command, an edit, a prompt. Across a whole engineering organisation that is thousands of actions a day, and today almost none of them are recorded.

    02 · TIMELINE

    cot puts them in order.

    Hooks in Claude Code, Cursor, Codex and other agents send each action to the collector. Every session becomes one timeline, normalised so every agent reads the same.

    03 · METRICS

    Then it counts what matters.

    Tool frequency, models, tokens, estimated cost and error rates, per session, per project and per team. See what your agents actually spend their time on.

    04 · INSIGHTS

    Detections find the few that need you.

    Secrets in context, edits outside the project, agents stuck in retry loops, cache paid for and never read. Each finding comes with evidence and a fix.

    05 · THE BIGGER PICTURE

    One session tells a story. Teams reveal the patterns.

    With a shared collector, teams can see where the same risks, retry loops and wasted spend keep showing up across projects. The actions you started with become evidence for what to improve next.

    What are your agents actually doing?

    Built-in detections read every session for security, cost and usability. They run locally, with no model calls, and they are free and open source.

    SECURITY

    See what agents can reach, and what they do with it.

    See behavior, not just configuration.

    COST

    Follow the money: tokens, models, retries and failed runs.

    Know what the work actually costs.

    USABILITY

    Find where people and agents lose time together.

    Improve the way people work with AI.

    security.sensitive_files

    Sensitive files accessed

    FOR EXAMPLE

    .env.production read into context

    Each finding comes with the evidence and the fix.

    for teams and organizations

    Understand AI usage across your whole organization.

    One shared collector for every engineer's agents. Each team sees its own sessions, and admins decide who sees what.

    Book a callSee pricing

    • Engineering leaders

      Which agents and models actually pay off?

      See usage, tokens and cost team by team. Spend spikes are flagged against your own baseline.

      • platform$1,240
      • payments$790
      • growth$410

    • Security teams

      What did agents read, run and send?

      Know what happened without traces leaving your machines. Detections cover secrets, sensitive files and exfiltration.

      READ.env.productionwarn

    • Platform teams

      How do we roll this out?

      One command. No SDK and no repo changes. Hooks wire into Claude Code, Cursor, Codex and OpenCode.

      $ curl -fsSL https://cot.run/install | sh
    Enterprise planeverything in open source, plus
    01Org-wide visibility
    Every engineer's agent sessions in shared team dashboards, searchable across the org.
    02SSO / SAML
    Sign in through the identity provider you already run.
    03Roles and access
    Decide who sees which teams, projects and traces.
    04Retention and audit
    Set how long traces live, and keep a record of who viewed or exported them.
    05Internal integrations
    Connect cot to the identity, alerting and data tools you already run.
    06Priority support
    Onboarding for your teams and a direct line to the people building cot.
    Organization settingsadmin console · illustrative

    identity

    retention

    teams and roles

    • SecurityAdmin
    • PlatformLead
    • PaymentsMember

    audit log

    • nowsecurityviewedsess_4c21e8
    • 2msystempurgedtraces older than 90d
    • 4mplatformopenedteam dashboard
    • 6madminchanged rolepayments → member

    get started

    Start with your own agents.

    One command. Self-hosted. Free and open source. See pricing

    Building with agents across a team?
    Talk to us about cot for organizations