What is code provenance?

Code provenance is the recorded origin and authorship of every line of code: who typed it, whether it was written manually, generated by an AI tool such as GitHub Copilot, Cursor, or Claude Code, or pasted from elsewhere. OmniType captures this provenance at write-time in the editor so each line carries verifiable authorship metadata.

Can OmniType detect AI-generated code?

Yes. OmniType detects when code is inserted by AI assistants like GitHub Copilot, Cursor, and Claude Code and labels each line with its source, giving teams an objective human-vs-AI breakdown per file, commit, and developer.

OmniType is free forever for individual developers, including the coding speed tracker and AI code detection in the VS Code extension. Paid plans add team analytics and large-scale hiring assessments.

What is OmniType Pulse?

OmniType Pulse is the HR side of the OmniType office suite: a multi-tenant HRMS for GPS-geofenced attendance, shift scheduling, leave, expenses, salary advances, and payroll with built-in Indian statutory compliance (PF, ESI, TDS).

OPEN STANDARD

ne file.
Every AI tool.

Before any AI-assisted file edit, write a single JSON file to disk. No SDK. No network. No account. Any tool writes it. Any reader consumes it.

<project>/.omnitype/active-model.json

{
  "model": "claude-sonnet-4-6",
  "tool":  "aider",
  "ts":    1747234567890,
  "file":  "/project/src/main.py"
}

Under 1ms

One atomic file write. No dependency, no API key, no network round-trip. Works in any language in 5 lines.

Your users get attribution

When your tool writes this file, every user gets automatic per-line AI attribution in OmniType and any reader of this spec.

Ships in 13 tools today

Claude Code ships it natively. Cursor, Windsurf, Codex, Cline, Copilot, Gemini CLI, Amp, and more auto-install it. Your users already benefit.

Confidence Tiers · v2

v2 grades every attributed span by how it was sourced. The higher the tier, the less the daemon has to guess. Emit the strongest signal your tool can.

T0Deterministic

Tool emits signed events with exact byte ranges via @omnitype-code/agent-sdk. No inference.

T1Hook-inferred

Tool writes the active-model.json sentinel before each edit (the spec below). Strong, zero-dependency.

T2Retrospective

Classifier reconstructs origin after the fact from timing and process ancestry.

T3Filesystem-only

A bare file change with no signal — best-effort guess.

13 TOOLS LIVE

Works where you already work

Auto-installed by the OmniType extension. Zero config on your end.

Claude Code

Anthropic

PreToolUse hook — ships natively

Cursor

Anysphere

hooks.json preToolUse — auto-installed

Windsurf

Codeium

pre_write_code hook — auto-installed

Codex

OpenAI

PreToolUse hook — auto-installed

Cline

PreToolUse script — auto-installed

Copilot

GitHub

~/.copilot/hooks/ PreToolUse

Gemini CLI

Google

settings.json BeforeTool — auto-installed

Droid

Factory

settings.json PreToolUse — auto-installed

Firebender

hooks.json preToolUse — auto-installed

Amp

TypeScript plugin — auto-installed

OpenCode

TypeScript plugin — auto-installed

TypeScript extension — auto-installed

Antigravity

Google

state.vscdb — auto-detected

Coming soon

Aider· no native hook API yetContinue· tracking upstreamGoose· pre-edit hook — proposed

Maintaining a coding tool?Get listed

Integration

Two ways in. For the strongest signal (T0), emit deterministic events from the SDK. For a zero-dependency path (T1), write the sentinel file in your preToolUse hook.

T0Deterministic SDKnpm i @omnitype-code/agent-sdk

import { OmniTypeAgent } from '@omnitype-code/agent-sdk';

const agent = await OmniTypeAgent.tryConnect({
  workspace, sessionId, tool: 'my-agent', model: 'gpt-4o',
});
const txn = await agent.beginTxn(['src/index.ts']);
// …apply the edit, recording exact byte ranges…
await agent.commitTxn(txn, edits);

T1Sentinel file— zero deps. Call this in your preToolUse hook. Pick your language.

import { writeFileSync, mkdirSync } from 'fs';
import { homedir, cwd } from 'os';
import { join } from 'path';
function signalOmniType(model: string, tool: string, file?: string) {
  const payload = JSON.stringify(
    Object.assign({ model, tool, ts: Date.now() }, file && { file })
  );
  for (const base of [cwd(), homedir()]) {
    try {
      const dir = join(base, '.omnitype');
      mkdirSync(dir, { recursive: true });
      writeFileSync(join(dir, 'active-model.json'), payload);
    } catch {}
  }
}
signalOmniType('gpt-4o', 'my-agent', '/project/src/index.ts');

Schema

All UTF-8. No nesting. No arrays.

fieldtypedescription

modelstring

Full model identifier. e.g. gpt-4o, claude-sonnet-4-6

toolstring

Stable tool slug in kebab-case. e.g. aider, cline

tsnumber

Unix epoch milliseconds at time of write.

filestring

Absolute path of the file about to be edited. Strongly recommended.

genIdstring

UUID v4 grouping multi-file edits in one agent turn.

required

optional

Rules

Write before the edit

The sentinel must land on disk before the file edit begins, not after.

Write atomically

Use a temp file and rename to avoid partial reads by the reader.

Write to both paths

Write to <project_root>/.omnitype/ (process.cwd()) AND ~/.omnitype/ so readers at both scopes get attribution.

Include the file field

Optional in the spec, but strongly recommended. Enables exact per-file attribution with no TTL race.

Skip if model unknown

Do not write a placeholder. If you cannot resolve the model, skip writing.

Stable. Forever free. No approval needed.

The file path is permanent. Fields only grow. No registration required. Write the file and you are compatible.

ne file.Every AI tool.

Works where you already work

ne file.
Every AI tool.