AI coding tools, measured across your whole engineering org.
LineTrail records which lines in commits and pull requests came from AI coding tools, with tool and model, and turns that into analytics across teams, repositories and pull requests. Evidence, not surveys or guesses.
Capture → Record → Sync → Measure
- PR #1842Squash-merged into acme/payments-api, attribution keptVerified
- PR #1839OpenCode · lines reported by the toolReported
- acme/data-pipelinesNo capture evidence for this repositoryUnknown
AI-attributed merged lines by team
| Team | Merged lines, by evidence | Verified | Reported |
|---|---|---|---|
| Payments | Verified 34%, Reported by tool 12%, Unknown, not estimated 6% | 34% | +12% |
| Checkout | Verified 22%, Reported by tool 9%, Unknown, not estimated 10% | 22% | +9% |
| Platform | Verified 41%, Reported by tool 7%, Unknown, not estimated 3% | 41% | +7% |
| Data | Unknown, not estimated | unknown | — |
- Verified
- Reported by tool
- Unknown, not estimated
01 The problem
AI coding spend is growing.Its impact is unmeasured.
AI coding tools now sit on a budget line, with seats, usage and cost you can count. What they contribute to your codebase and delivery is much harder to see: by team, by repository, by pull request. Surveys and guesses don’t hold up when someone asks where the investment pays off.
On the invoice
- Seats
- Usage
- Cost
Missing from the report
- Which teams AI-written code lands in
- Which repositories and pull requests
- Which tools and models produced it
- Whether a figure is verified, reported or unknown
Facts, not projections
- Evidence
- 3
- Evidence levels on attribution: verified, reported by tool, and unknown.
- Integrations
- 6
- Integrations available today across AI tools, source control and VS Code.
- Source control
- 3
- Source control platforms, including Bitbucket Server / Data Center, through read-only connectors.
- Access
- 3
- Roles (admin, member, viewer), with SSO via OIDC.
02 How it works
From commit to dashboard,one continuous trail.
Attribution is recorded as Git notes beside the commits it describes, kept through squash and rebase merges, and turned into analytics on your own infrastructure.
- 01 / 04
Capture
A CLI and capture adapter record which lines came from an AI coding tool, with tool and model.
- 02 / 04
Record
Attribution is stored as Git notes beside each commit, and kept through squash and rebase merges.
- 03 / 04
Sync
Your self-hosted LineTrail server ingests the notes.
- 04 / 04
Measure
The web dashboard shows repositories, pull requests, teams and contributors, tools and models, usage and cost.
VS Code, line by line.
The VS Code extension shows attribution in the gutter and on hover, with branch changes and history views.
03 Analytics
One view of AI contribution,across the org.
See where AI coding tools contribute, broken down the way your organization works: repositories, pull requests, teams and contributors, tools and models.
Repositories
| Repository | Merged PRs | Merged lines, by evidence | Verified |
|---|---|---|---|
| acme/payments-api | 214 | Verified 38%, Reported by tool 10%, Unknown, not estimated 4% | 38% |
| acme/checkout-web | 167 | Verified 20%, Reported by tool 11%, Unknown, not estimated 9% | 20% |
| acme/platform-infra | 122 | Verified 44%, Reported by tool 6%, Unknown, not estimated 2% | 44% |
| acme/data-pipelines | 58 | Verified 0%, Reported by tool 0%, Unknown, not estimated 100% | unknown |
Pull requests
| Pull request | Tool · model | AI share |
|---|---|---|
| #1842 Idempotency keys for retriesacme/payments-api | Claude Code · Sonnet | 62% (Verified) |
| #1839 Split ledger writeracme/payments-api | OpenCode | 18% (Reported by tool) |
| #977 Cart totals roundingacme/checkout-web | Claude Code · Opus | 41% (Verified) |
| #412 Backfill job for eventsacme/data-pipelines | unknown | unknown |
Teams and contributors
| Team | Contributor |
|---|---|
| Payments 14 contributors | Hidden for viewers |
| Checkout 11 contributors | Hidden for viewers |
| Platform 9 contributors | Hidden for viewers |
Tools and models
| Tool · model | Merged lines, by evidence | AI lines, verified |
|---|---|---|
| Claude Code · Sonnet | Verified 30%, Reported by tool 9% | 18,240 |
| Claude Code · Opus | Verified 12%, Reported by tool 4% | 7,115 |
| OpenCode | Verified 8%, Reported by tool 6% | 5,980 |
- Verified
- Reported by tool
- Unknown, not estimated
04 Usage and cost
Usage and cost,next to contribution.
Put what AI coding tools cost beside what they contributed, by tool, model and team, so the budget conversation starts from evidence.
Usage and cost that a tool reports are marked as reported, so they stay distinct from verified figures.
| Tool · model | Team | Sessions | Cost | AI lines merged, verified |
|---|---|---|---|---|
| Claude Code · Sonnet | Payments | 1,204 (Reported by tool) | $2,310 (Reported by tool) | 9,860 |
| Claude Code · Opus | Platform | 386 (Reported by tool) | $1,870 (Reported by tool) | 6,420 |
| OpenCode | Checkout | 742 (Reported by tool) | $640 (Reported by tool) | 3,105 |
| Claude Code · Sonnet | Mobile | 511 (Reported by tool) | $980 (Reported by tool) | unknown |
- Verified
- Reported by tool
- Unknown, not estimated
05 Integrations
What LineTrailconnects to.
What’s available today, and what’s coming soon. Source control connectors are read-only.
- Claude CodeAI coding toolAvailable
- OpenCodeAI coding toolAvailable
- CodexAI coding toolComing soon
- CursorAI coding toolComing soon
- GitHub CopilotAI coding toolComing soon
- GeminiAI coding toolComing soon
- GitHub / GitHub EnterpriseSource control, read-only connectorAvailable
- GitLabSource control, read-only connectorAvailable
- Bitbucket Server / Data CenterSource control, read-only connectorAvailable
- VS CodeEditorAvailable
- JetBrains IDEsEditorComing soon
- JenkinsDelivery outcomesComing soon
- SonarQubeQuality outcomesComing soon
- AI InsightsOver your dataComing soon
- AI ChatOver your dataComing soon
06 Security and control
Enterprise controls.Private by default.
SSO, roles and scoped tokens for your organization, read-only connectors for the source control you run, and self-hosted deployment today.
Controls
- Deployment
- Self-hosted today. Managed cloud coming soon.
- Source control
- GitHub / GitHub Enterprise, GitLab, and Bitbucket Server / Data Center, through read-only connectors.
- Single sign-on
- SSO via OIDC.
- Roles
- Admin, member and viewer roles.
- Contributor privacy
- Contributor identity is hidden from viewers.
- API access
- Scoped API tokens.
- Tenancy
- Organizations.
Data
LineTrail is metadata-first. Raw source code, prompts and transcripts are not sent to a central service by default.
Recorded and synced
- Attribution metadata as Git notes: which lines, which tool, which model
- Ingested by your self-hosted LineTrail server
Not sent to a central service by default
- Raw source code
- Prompts
- Transcripts
Principle, not a testimonial
“When evidence is missing, the number says unknown. Never a guess.”
Numbers you can defend.
LineTrail separates what a tool reported from what was verified. When evidence is missing, it shows “unknown” instead of guessing. When someone asks where a number came from, you have an answer.
Verified
Attribution that was verified. Kept apart from anything a tool only reported.
Reported
What an AI coding tool reported. Shown, and always labeled as reported.
Unknown
No evidence, no estimate. The figure reads “unknown” rather than a guess.
Questions, answered plainly.
Does our code leave our network?
LineTrail is self-hosted and metadata-first. Raw source code, prompts and transcripts are not sent to a central service by default. Optional features are explicit opt-in, with access controls and retention.
Which tools and source control systems are supported?
Available now: Claude Code and OpenCode; GitHub / GitHub Enterprise, GitLab and Bitbucket Server / Data Center through read-only connectors; and VS Code. Coming soon: Codex, Cursor, GitHub Copilot and Gemini; JetBrains IDEs; Jenkins and SonarQube delivery and quality outcomes; and AI Insights and AI Chat over your data.
What happens with squash and rebase merges?
Attribution is recorded as Git notes and kept through squash and rebase merges, so rewritten history doesn’t erase it.
How accurate is it?
LineTrail doesn’t promise perfect attribution. It separates what a tool reported from what was verified, and when evidence is missing it shows “unknown” instead of guessing. Reported and verified figures stay distinct.
Request early access.
LineTrail is in early access. Tell us about your team.
Early access · No public release yet