See what your local AI coding tools are doing before the invoice, the incident, or the runaway session surprises you. AIWatcher Local is the open-source individual tier: a private, zero-code-change view of Claude Code, Codex CLI, and Cursor activity on your machine. No account.
~/code/ml-pipeline crossed 250 model calls in one session — add an explicit stop condition before running it again.
| Tool | Project | Tokens | API value | Outcome |
|---|---|---|---|---|
| claude-code | ~/code/api-gateway | 551.9k | $8.08 | Useful |
| codex-cli | ~/code/ml-pipeline | 179.1M | $0.00 | Rework |
| claude-code | ~/code/data-sync | 330.9k | $4.93 | Useful |
| claude-code | ~/code/web-app | 127.4k | $1.72 | Useful |
| claude-code | ~/code/meeting-notes | 59.6k | $0.65 | — |
| claude-code | ~/code/api-gateway | 20.9k | $0.31 | Useful |
Enterprise adds team history, policy controls, HITL approvals, evidence packs, and integrations.
One developer, one machine, one read-only view. It answers the questions you actually ask when the bill looks off or a session ran longer than you meant it to:
Which local AI coding tools are active right now.
Which projects are driving usage, down to the git project root, not a parent folder.
Which models are generating the most API-equivalent value.
Which usage is API-priced versus subscription or plan-limited.
Which recent sessions are worth inspecting, with click-through detail.
What happened inside a session, without showing prompt or source text.
Which prompt or turn inside a session actually drove the cost, and which one is worth tightening next time.
Most tools tell you what an AI session cost after it’s over. AIWatcher Local can step in before a risky prompt runs at all.
Every prompt gets a local risk score — broad, destructive, vague, or expensive work is flagged before execution. AIWatcher then produces an intent-preserving execution brief: the same request, scoped down to what’s actually needed.
Install a native hook for Claude Code or Codex CLI once, and every prompt is preflighted automatically — no workflow change.
For anything medium or high risk, a local decision screen (nothing leaves your machine) gives you four options: use the brief, edit it first, run the original anyway, or cancel. Low-risk prompts pass straight through.
Works even where hooks can’t reach: Claude Desktop, web chat, and editor sidebars don’t expose hook points yet. The dashboard’s Prompt Companion gives you the same preflight check as a copy-paste widget.
Turn it on:
$ aiwatcher install-claude-hook --write --scope user --gate
$ aiwatcher install-codex-hook --write --scope user --gate
The fast terminal version. A daily summary of measured AI work, broken down by tool, model, and project.
A local dashboard at 127.0.0.1:8765. Served from the standard library, no external assets. Project rows and sessions are clickable.
And here is a local session in the terminal. Scroll for the weekly report and per-project projects breakdown:
$ aiwatcher today
Today - Tuesday, June 23, 2026
3 sessions · 551.9k API-priced tokens · 239.0M plan/limited observed · $8.08 API-equivalent value
Projected month: ~$71.09 API-equivalent at current pace
Note: subscription plans may not bill this as incremental spend.
By tool
Tool API value Calls Tokens Sessions
-------------------------------------------------------
claude-code $8.08 896 551.9k 1
codex-cli $0.00 2 239.0M 2
By model
Model API value Tokens Calls
-------------------------------------------------
claude-sonnet-4-6 $8.08 551.9k 896
gpt-5.5 $0.00 239.0M 2
Top project: ~/code/api-gateway (100% of today's API-equivalent value)
Note: Codex cost is subscription/plan-based, not estimated as API spend
This week: $15.72 This month: $54.50
$ aiwatcher report --days 7
AIWatcher Local report - last 7 days
Sessions: 9 · API-equivalent value: $15.72 · Tokens: 358.2M
Model calls: 1888 · Tool calls: 1165
Top project: ~/code/api-gateway ($8.39)
Top tool: claude-code (6 sessions)
Top model: claude-sonnet-4-6 (1.1M tokens)
Suggested next checks:
- Review top project sessions for runaway or abandoned work.
- Compare API-priced tokens with plan/limited tokens before reading invoice impact.
- Run `aiwatcher ui` for clickable local drill-down.
$ aiwatcher projects --days 7
AI usage by project - last 7 days
Cost is shown as API-equivalent value; subscription plans may differ.
$8.39 2 sessions 576.0k tokens ~/code/api-gateway
$4.93 1 session 330.9k tokens ~/code/data-sync
$1.72 1 session 127.4k tokens ~/code/web-app
$0.67 2 sessions 62.5k tokens ~/code/meeting-notes
The dashboard exposes a read-only JSON summary at /api/summary. The time-window selector updates the values, and project rows and recent sessions open into local drill-down. It is the first thing you would screenshot and show a teammate.
Local trust boundary
If AIWatcher Local cannot explain what it reads and why, it should not read it.
Reads ~/.claude/projects/**/*.jsonl, uses event-level working directory, and normalizes nested folders to the git project root when possible. The richest support today.
Reads ~/.codex/state_5.sqlite in read-only mode when available. Support depends on local SQLite accessibility and schema shape.
Detects local AI log activity where available. Cost and token detail is marked limited on purpose, because Cursor does not reliably expose it locally.
Detected when installed. Full session scanning isn't implemented yet — AIWatcher Local confirms Cline is present but doesn't extract session or cost detail from it today.
Detected when installed. Full session scanning isn't implemented yet — AIWatcher Local confirms Windsurf is present but doesn't extract session or cost detail from it today.
Event exports include timestamps, models, token counts, API-equivalent value, project paths, event types, and content hashes. They do not include prompt text or source code.
Mark any session useful, rework, or abandoned with aiwatcher outcome — the record that closes the loop between what you did and what it was worth.
AIWatcher Local answers
What are my local AI coding tools doing, where is usage growing, and what is the API-equivalent value before the invoice arrives?
The paid tier answers
What is our entire team's AI work doing, who is driving the biggest costs, and where are the risks before they become incidents?
AIWatcher Local is never gated behind signup. The paid tier builds on the same collector with team rollups, anomaly detection, cost attribution by team and customer, and integrations for the workflows engineering leaders already use.
AIWatcher Local ships as a separate package you run on your own machine. No account, no signup — just install it.