Everyone is building agents.Who's keeping track?
Octic keeps one live map of every AI agent, MCP server, coding agent, and package you run, built from your APIs, your network traffic, and sensors on each host. Every one gets an owner and a trust state.
Found below the API. An API-only inventory never lists it.
The map, in cross-section
Every AI thing at a sample company, by the layer that found it: API, network, or a sensor on the host.
API connections5
Network traffic+3
Under the surface, 1 more. Choose + network to look.
Kernel sensors+8
Under the surface, 6 more. Choose + kernel to look.
Find the AI that API-only tools miss
Octic reads your cloud and SaaS APIs, watches network traffic, and runs kernel-level sensors on Linux and Windows hosts. Each layer finds AI the one above it can't see, and all of it lands on one map.
- API connections5 foundsupport-triage-agentmcp-jiraAnthropic APIOpenAI APIpayments-api
- Network traffic+3 foundinvoice-extractormcp-postgres-prodmcp-filesystem
- Kernel sensors+8 foundeng-ws-14WIN-ENG-07ip-172-31-67-241Claude CodeCodexAnthropic SDK 1.3.0openai 1.51.0shadow-summarizer
- An API-only inventory stops at 5.16 of 16 on the map
Decide what's trusted, and who owns it
Octic-managed policies with enhanced protections apply from day one, and you can add your own. An ownership engine works out who owns each node, level by level. Anything it can't place shows up with no owner.
10approved3unreviewed3flagged
- CodexFlaggedSigned in with a personal accountNo owner
- shadow-summarizerFlaggedAn AI agent nobody ownsNo owner
- mcp-postgres-prodFlaggedMCP server with production database accessNo owner
- openai 1.51.0UnreviewedModel SDK nobody has reviewedNo owner
- invoice-extractorUnreviewedReads invoices, awaiting review@finance-ops
- mcp-filesystemUnreviewedMCP server reading local files@platform
4 of these 6 have no owner. The other 10 nodes are approved and owned.
What did $3,000 of AI coding buy?
Agentic Coding Insights answers from the same sensors. On every host with one, Claude Code and Codex sessions show up on their own: who ran them, the prompt, the repo, and the cost, split by kind of work. More coding agents are coming, and so is the Octic SDK, which will report on the AI agents you build into your own apps.
What Priya's AI coding bought
Priya Raman · Payments · eng-ws-14
| Kind of work | Share | Spend |
|---|---|---|
| Feature work | $1,684.20 | |
| Refactors | $512.10 | |
| Tests | $401.30 | |
| Docs | $236.80 | |
| Debugging | $178.00 |
Add idempotency keys to the refund endpoint and cover retries
Claude Codepayments-apiFeature work$41.80Split ledger.ts into posting and reconciliation modules
Claude Codepayments-apiRefactors$22.15Write integration tests for the checkout currency switcher
Codexpayments-webTests$12.40Document the webhook signature verification flow
Claude Codepayments-apiDocs$6.90
Why is the staging plan trying to recreate the NAT gateway?
Codexinfra-terraformDebugging$9.35FlaggedPersonal account, not a corporate seat
Every risk gets an owner and a due date
Each risk becomes a finding with an owner, an SLA, and an event history you can export as CSV. An exception needs an admin's approval, and it expires.
Coming late 2026, Octic will draft the fix and name who should approve it. Nothing will change until that person says yes.
- Status
- Awaiting approval
- Proposal
- rem_142
- Risk
- Codex session on a personal account
- Resource
- WIN-ENG-07 · Marcus Webb
- Proposed by
- Octic
- Approver
- Lena Okafor · @platform-leads
Preview only. When agentic remediation ships, Lena will approve or decline here.
Start from your question
One map, three ways in. Pick the one that sounds like you.
Engineering leaders: what is our Claude Code and Codex spend buying?
Sessions, prompts, and spend per engineer, split by kind of work: feature work, refactors, tests, docs, debugging, and more.
On the map: Claude CodeCodexeng-ws-14WIN-ENG-07payments-api
Security leaders: is anyone on a personal Codex account?
Every AI agent, MCP server, and package, including the ones only a network or kernel sensor can see, with its trust state.
On the map: Codexshadow-summarizermcp-postgres-prodopenai 1.51.0mcp-filesystem
Operators and COOs: who owns each AI tool, and what does it cost?
An owner for every node, the model APIs behind them, and coding spend you can read without a dashboard tour.
On the map: Anthropic APIOpenAI APIClaude Codesupport-triage-agentinvoice-extractor
Put your own estate on the map
Install a sensor on a Linux or Windows host and coding-agent sessions start showing up, with no extra setup. Connect your APIs and the rest of the map fills in.