See current code contribution by AI tool and model, inspect observed repository token activity, and keep evidence gaps visible—all through a local CLI with no account or telemetry.
Fictional repository data. Contribution describes the current code snapshot; token activity covers observed work over time.
1Why LLM Status
A clearer picture of a codebase shaped by many tools.
One repository, several AI tools
Keep the editing tool and the reported model separate, so a multi-tool workflow does not collapse into one generic AI bucket.
Current lines by tool and model
Contribution that reflects current code
See the ownership evidence for Git-tracked lines that survive today instead of treating historical edit counts as the repository snapshot.
AI · Measured Human · Mixed · Unknown · Unmeasured
Token visibility beyond file changes
Inspect observed repository activity over a selected period while keeping incomplete counters and coverage gaps explicit.
Input · output · total · cache · reasoning
2Fictional product example
Current contribution and observed activity, kept distinct.
This entire example is fictional. The same shared dataset drives every number below.
Selected period · Partial token coverage
Illustrative offline-report mockup—not a live repository screenshot.
Ownership · 10,000 tracked lines
AI6,000 · 60%
Measured Human2,500 · 25%
Mixed500 · 5%
Unknown400 · 4%
Unmeasured600 · 6%
Current AI contribution
Surviving lines; percentages use all tracked lines.
Tool
Recorded model
Lines
Repository
Codex
gpt-5
2,000
20%
Cursor
claude-sonnet-4
1,600
16%
Claude Code
claude-sonnet-4
1,200
12%
Gemini CLI
Unknown model
600
6%
Kimi Code CLI
Unknown model
400
4%
DeepSeek Harness
Unknown model
200
2%
Identifiable-model coverage: 4,800 of 6,000 AI-owned lines · 80%
Token usage covers observed activity over time. Code contribution describes the current repository snapshot. Their ratio is not an efficiency, ROI, productivity, or model-quality score.
3How it works
Evidence that becomes more useful as you keep it running.
01
Install and initialise
Install the public npm package globally, then initialise LLM Status inside an existing Git repository.
02
Connect tools and activate capture
Enable one or more qualified integrations, follow each activation instruction, and start a fresh trusted session.
03
Inspect as evidence grows
Scan current contribution, open the offline report, and optionally refresh supported token activity while you work.
4Qualified integrations
Tool support without hiding the boundaries.
Editing-tool capture, per-edit model evidence, and token collection are separate support surfaces.
Tool / surface
Editing evidence
Model evidence
Token collection
Qualified setup
Claude CodeCLI · VS Code extension
Write / Edit hooks
When the file hook supplies it; standard hooks often omit it
Native project-session sync
2.1.139+ · trusted project hooks
CodexCLI · local desktop · IDE conditional
Structured apply_patch hooks
When the file hook supplies it
Native local rollout sync
CLI 0.124.0+ · trusted project hooks
CursorIDE
Agent Chat / Cmd+K and Tab edit hooks
Explicit model_id or model from the file hook
External normalized import only
1.7.0+ · trusted workspace hooks
Gemini CLICLI · VS Code Companion
write_file / replace hooks
Unknown with the current adapter
Native project-chat sync
Stable 0.58.0+ within 0.x · trusted project hooks
Kimi Code CLICLI · IDE via ACP
Write / Edit plugin callbacks
Unknown with the current adapter
Native wire 1.5 journal sync
Stable 0.41.0+ within 0.x · plugin activation required
DeepSeek HarnessCLI · local web · IDE via ACP
Native write / edit plugin events
Unknown with the native plugin
Native session sync
dsh 0.1.2-rc.1 preview · launch overlay required
Claude Code
CLI · VS Code extension
Editing evidence
Write / Edit hooks
Model evidence
When the file hook supplies it; standard hooks often omit it
Token collection
Native project-session sync
Qualified setup
2.1.139+ · trusted project hooks
Codex
CLI · local desktop · IDE conditional
Editing evidence
Structured apply_patch hooks
Model evidence
When the file hook supplies it
Token collection
Native local rollout sync
Qualified setup
CLI 0.124.0+ · trusted project hooks
Cursor
IDE
Editing evidence
Agent Chat / Cmd+K and Tab edit hooks
Model evidence
Explicit model_id or model from the file hook
Token collection
External normalized import only
Qualified setup
1.7.0+ · trusted workspace hooks
Gemini CLI
CLI · VS Code Companion
Editing evidence
write_file / replace hooks
Model evidence
Unknown with the current adapter
Token collection
Native project-chat sync
Qualified setup
Stable 0.58.0+ within 0.x · trusted project hooks
Kimi Code CLI
CLI · IDE via ACP
Editing evidence
Write / Edit plugin callbacks
Model evidence
Unknown with the current adapter
Token collection
Native wire 1.5 journal sync
Qualified setup
Stable 0.41.0+ within 0.x · plugin activation required
DeepSeek Harness
CLI · local web · IDE via ACP
Editing evidence
Native write / edit plugin events
Model evidence
Unknown with the native plugin
Token collection
Native session sync
Qualified setup
dsh 0.1.2-rc.1 preview · launch overlay required
Copilot, Windsurf, arbitrary editor extensions, generic clipboard changes, and cloud-agent capture do not currently have automatic adapters. Explicit manual declarations remain user-supplied evidence.
5Ongoing attribution
Model changes do not rewrite the past.
A one-time scan can only use the evidence available now. Keep supported integrations active so new edits can add better evidence without relabelling untouched lines.
MODEL A100 lines
→
MODEL A80
MODEL B20
If model B explicitly replaces 20 lines, the current snapshot can show 80 for A and 20 for B. Untouched lines keep their earlier evidence. Unknown model remains distinct from Unknown or Unmeasured code origin.
6Local by design
Evidence stays close to the repository.
The CLI makes no network calls and collects no telemetry. It does not store raw source, prompts, responses, tokens, or credentials in its provenance ledger or usage store.
Private metadata can still be sensitive: paths, tool/model/session IDs, hashes, and local installation paths are not anonymous. Review private artifacts before sharing.
Supported local evidence
→
Local CLI
→
Terminal / offline report
Optional sharing path
Aggregate summary
→
README badge
7Quick start
Start in an existing Git repository.
Requirements: Node.js 20+, Git 2.45+, and an existing Git repository. A stable global installation is recommended for ongoing hooks.
$npm install -g llmstatus@latest
$cd your-repository
$llmstatus init
$llmstatus integrate codex
Follow the printed activation instructions, start a fresh trusted session, and make a supported edit to a Git-tracked file. Multiple integrations can be enabled in the same repository.
Inspect results
$llmstatus scan
$llmstatus report
$llmstatus badge --write
Optional token collection
Token collection is off by default, opt-in per checkout, command-triggered under a cooldown, and never a background daemon.
$llmstatus usage auto enable
$llmstatus usage
Automatic capture is currently qualified primarily for main Git worktrees on macOS and Linux. Linked worktrees are unsupported, and live Windows integration qualification is incomplete.
8Frequently asked
Questions, answered.
9Get started
Start with your next coding session.
Install the public MIT-licensed package, connect a supported tool, and let reliable evidence accumulate from here.
Illustrative example