Skip to content
Want to see what AI is doing?Try Octic for free
octic.ai

One install. See what AI does.Know what the spend bought.

Octic runs on your laptops and servers and maps the AI agents, MCP servers, and coding sessions on each host. It splits coding spend by intent, like feature work or tests.

One command. No code changes.

$ curl -fsSL https://<workspace>.octic.ai/install | sh

octic-sensor installed · macOS 15.1

Connected to Acme Corp

Watching Claude Code, Codex, MCP servers

Acme Corp · ActivityLive

AI activity

Agents, MCP servers, coding sessions

  • Add idempotency keys to the refund endpoint and cover retriesPriya Raman · eng-ws-14Feature work$41.80
  • MCP server mcp-jiraOpened an MCP session to mcp-jirasupport-triage-agent · ip-172-31-67-241MCP call—
  • Split ledger.ts into posting and reconciliation modulesPriya Raman · eng-ws-14Refactors$22.15
  • Write integration tests for the checkout currency switcherPriya Raman · eng-ws-14Tests$12.40
  • Called gpt-4.1 to read an invoice PDFinvoice-extractor · 172.31.40.18Model call$0.04
  • Document the webhook signature verification flowPriya Raman · eng-ws-14Docs$6.90
  • Why is the staging plan trying to recreate the NAT gateway?Marcus Webb · WIN-ENG-07Debugging$9.35
  • Called Claude to route a support ticketsupport-triage-agent · ip-172-31-67-241Model call$0.11

Integrations and platforms that ship today

  • OpenAI
  • Anthropic
  • Claude Code
  • Codex
  • AWS
  • Microsoft
  • GitHub
  • macOS
  • Windows
  • Linux

One sensor per host. That's the setup.

Nothing to configure per app or per repo. Each sensor reports what AI does on its host.

  1. 1
    macOSWindowsLinux

    $ curl -fsSL https://<workspace>.octic.ai/install | sh

    Install the sensor

    Run one command on each laptop or server, on macOS, Windows, or Linux. Nothing to change in your code.

  2. 2
    • Claude Codepayments-api · Feature work
    • Codexinfra-terraform · Debugging
    • MCPmcp-jirasupport-triage-agent · MCP call

    AI activity shows up

    Octic finds the AI agents and MCP servers on each host. Claude Code and Codex sessions arrive with engineer, prompt, repo, and model.

  3. 3
    • Feature work44%
    • Refactors19%
    • Debugging14%
    • Tests14%

    Measure spend against intent

    Octic tags each session with its intent and its cost. Spend then rolls up by the work it paid for.

What did $3,000 of AI coding buy?

Intent is the kind of work a session was doing. Octic rolls spend up by intent, engineer, repo, and model, from a month of work down to one session.

  • By intentFeature work, refactors, tests, docs, debugging, and more, each with its spend and session count.
  • By engineer, repo, and modelSee who drove spend, in which repo, on which model, and how it moved week over week.
  • Cost per sessionOpen any session to see its cost split across input, output, and cache tokens.
See Agentic Coding Insights
Where the money goes · Priya Raman

Coding agents · September 2026 · $3,012.40 across 214 sessions

  • Feature work$1,684.20
  • Refactors$512.10
  • Tests$401.30
  • Docs$236.80
  • Debugging$178.00

Feature work took 56% of Priya's $3,012.40.

Priya's recent sessions

  • Add idempotency keys to the refund endpoint and cover retriesClaude Codepayments-apiFeature work$41.80
  • Split ledger.ts into posting and reconciliation modulesClaude Codepayments-apiRefactors$22.15
  • Write integration tests for the checkout currency switcherCodexpayments-webTests$12.40

Spend is one view of the control plane

The same graph maps your AI agents and MCP servers, decides which are trusted, and sends what's wrong to an owner.

Discover. A live map of your AI agents and MCP servers

Provider APIs, network traffic, and kernel-level sensors find AI agents, MCP servers, API endpoints, packages, and processes.

How discovery works
  • support-triage-agentApproved@support-eng
  • mcp-filesystemUnreviewed@platform
  • mcp-postgres-prodFlaggedNo owner

Trust. Every AI agent and MCP server gets a trust state

Each starts unreviewed. Built-in and custom policies mark it approved or flagged. Ownership traces from repo to team to person.

How trust works

Observe. Each coding session, turn by turn

Prompts, tool calls, latency, and cost for Claude Code and Codex sessions. SDK tracing for your own AI agents is coming soon.

See a coding session
MCP server with production database accessFlagged
Assignee
@platform
SLA
3 days left
Resource
mcp-postgres-prod
  1. 09:12:21Detected: shadow-summarizer connected from eng-ws-14
  2. 09:12:24Assigned to @platform, SLA started
  3. 10:02:41Exception requested, waiting on an admin

Coming late 2026Octic-drafted fixes

Remediate. What's wrong becomes a finding with an owner

Each finding gets an SLA and an event history today. Octic-drafted fixes, approved by a person, are coming late 2026.

How findings work

Sensors see what cloud APIs can't

API integrations inventory projects, keys, models, and repos. Sensors add the AI apps, packages, processes, and coding sessions on each host, and the MCP servers they connect to.

  • OpenAI
  • Anthropic
  • Claude Code
  • Codex
  • AWS
  • Microsoft
  • GitHub
Octic
  • macOS sensorEngineer laptops
  • Windows sensorLaptops and servers
  • Linux sensorServers and dev machines

OpenAI, Anthropic, Claude, Codex, AWS, Microsoft, GitHub, Apple, macOS, Windows, and Linux are trademarks of their respective owners. Octic is not affiliated with or endorsed by them.

Start with one laptop

Sign up and run one command. Coding sessions arrive with cost and intent attached.

Design partners plan the roadmap with us. Join the Design Partner Network.

One command. No code changes.

$ curl -fsSL https://<workspace>.octic.ai/install | sh

octic-sensor installed · macOS 15.1

Connected to Acme Corp

Watching Claude Code, Codex, MCP servers