AI tool provenance and spend governance

Every prompt, every PR, every dollar.

Ink records every prompt, diff, and dollar from your coding agents, then rolls it up across your team and your org. Install once. AI spend governance based on outcomes. Know what you're getting for your money.

curl -fsSL https://raw.githubusercontent.com/Exceeds-AI/exceeds-ink-downloads/latest/install.sh | sh

How to get started ↓
capturesClaude CodeCursorCodexCopilotWindsurf

In the dashboards

The questions Ink lets you ask of your AI engineering.

Conversion funnel

Where does AI work fall off?

Track every generated edit from prompt to merged PR. Spot the stage that loses the most lines.

Generated
Committed
Opened in PR
Merged
1.9K
1.4K
887
799
100%
71%
47%
42%

Model mix

Which models are eating spend?

Share of total AI dollars by model, agent, and provider.

Sonnet 4.640%
Opus 4.825%
GPT-5.518%
Composer 2.512%
Fable 55%

Live totals

What does today look like?

Spend, commits, hours, and turns — pulled from every tool.

Spend
$4.4k
Commits
484
Hours
1,072h
Turns
8.5K

Spend allocation

Which features cost the most?

Roll up by feature, repo, or squad. Find the rework hotspots before they compound.

Checkout & payments$1,840 · 42%
Onboarding flow$1,094 · 25%
Search & ranking$831 · 19%
Reporting exports$219 · 5%
Admin dashboard$175 · 4%
Notifications$131 · 3%
Internal tools$87 · 2%

Governance (in beta)

One policy. Every agent obeys.

Admins define the rules in one place. Ink enforces them on every machine, inside every agent, at the moment of use.

Cap spend per engineer, team, or org. Soft warnings before hard stops.

Approve providers and models once. Unapproved ones never run.

Adjust models and allowances based on an engineer's history in orchestration, code quality, and other Exceeds-tracked signals.

Block traffic to endpoints you have not vetted, per repo or org-wide.

Write policies in plain language. Ink turns them into rules and enforces them like any other.

claude — acme/api

$ claude

policy.checkmodel=claude-fable-5 approved

tool.invokeedit app/billing/retry.ts +64 −12

budget.noticeteam=platform $412 of $500 weekly cap

budget.stopcap reached new sessions paused until Mon

session.commitsha=41c9f2 in-flight work lands safely

Getting started

One command and you're live.

  1. Run the install command

    Paste it into any terminal. Nothing to configure, no agents to touch.

    curl -fsSL https://raw.githubusercontent.com/Exceeds-AI/exceeds-ink-downloads/latest/install.sh | sh

    Verify it anytime with exceeds-ink verify.

  2. Connect your account

    The installer opens web.exceeds.ai in your browser to link this machine. No account yet? You create one right there.

  3. Your 7-day org trial begins

    Free for the whole org. Data starts flowing within about 15 minutes.

Installing at the org level? Silent install for IT teams

What Ink does

From your terminal to the boardroom. Same data, different lens.

Passive capture
no plugins, no workflow changes
Provenance in Git
every line tagged with agent, model, prompt
Private by default
prompts stay local until you say otherwise
Spend per agent
model, mode, and session
Skills that travel
your best prompts, packaged and shared
Rollups
engineer, team, org

Per-agent capture

Different agents. The same normalized stream.

For the engineer: every prompt, edit, and mode change, tied back to the exact lines that shipped.

Claude Code

exact capturedeterministic

Complete session capture with exact token counts and costs. Every prompt, mode switch, and diff deterministically attributed to the lines it produced.

session.start/endcaptured
tool.invokeedit · bash · read
tokensinput + output, per turn
costexact, per model
modesplan · agent · edit · headless
skills detectedyes

15 minutes

Install once. Provenance on every commit.

One command. No agents to reconfigure. Ink attributes every AI-written line to its agent, model, and prompt, starting with your next commit.

You

Your sessions. Your skills.

Every line you ship, traced to the prompt that wrote it. A personal record of your sessions, plus a prompting coach that runs in your own agent.

Team

Patterns that move across the team.

See which prompts produce code that survives review. Roll up by team or repo. Transfer skills from your strongest engineers to the rest of the team.

Org

Adoption, cost, and compliance.

Spend per agent, adoption per team, AI-authored percentage per repo. One number for the board. A privacy posture your CISO can verify.

Security, privacy, governance

Code on the machine. Prompts on the laptop. Notes you can verify.

Local-first by default

Source code, prompts, and transcripts stay on the developer's machine until you turn on rollups.

Signed, revocable ingest

When data flows, it carries an HMAC-SHA256 signature using a machine token you can revoke in one click.

Policy switches at install

Aggregate-only capture, per-repo allowlists, retention windows, server-side prompt redaction. Set once, audit anytime.

Deployment

Local. Self-hosted. Or hosted by Exceeds.

Bring your own Postgres or use ours. Same binary, same Notes, same dashboards — your choice on where they live.

Data points

Hundreds of metrics, pulled from one stream.

Every prompt, diff, mode, model, and cost — joined, normalized, and ready for the question you haven't asked yet.

Engineers see which prompt wrote each line. Managers see which patterns work. Execs see where the AI dollars went.

Total Tokens per Model
Cost per Repo per Week
Lines AI-Authored
Mode-mix by Engineer
Acceptance Rate (Copilot)
Sessions per Day
Tool Invocations per Session
Spend per Agent
Plan Turn Frequency
Skills Detected per Repo
Total Tokens per Model
Cost per Repo per Week
Lines AI-Authored
Mode-mix by Engineer
Acceptance Rate (Copilot)
Sessions per Day
Tool Invocations per Session
Spend per Agent
Plan Turn Frequency
Skills Detected per Repo
Tokens per Prompt
Files Touched per Session
Diff Confidence Score
Adoption Curve by Team
Sanitization Hits per Day
Notes Pushed per Day
Rollup Sync Latency
Per-Repo Allowlist Hits
Cost per Mode
Skill Transfer Success
Tokens per Prompt
Files Touched per Session
Diff Confidence Score
Adoption Curve by Team
Sanitization Hits per Day
Notes Pushed per Day
Rollup Sync Latency
Per-Repo Allowlist Hits
Cost per Mode
Skill Transfer Success

For IT teams

Rolling out to the whole org?

Deploy silently through Jamf, Intune, or any MDM. Non-interactive install with no user prompts and no reboots. Engineers just see their data show up.

curl -fsSL https://raw.githubusercontent.com/Exceeds-AI/exceeds-ink-downloads/latest/install.sh | sh -s -- --silent

Runs unattended on managed machines. SSO, SCIM, and audit controls are covered in the deployment guide.

Ink · public preview

Run Ink on your machine in 15 minutes.

One command. No agents to reconfigure. No data leaves your laptop without your say-so.

curl -fsSL https://raw.githubusercontent.com/Exceeds-AI/exceeds-ink-downloads/latest/install.sh | sh

Prefer a walkthrough first? Request a demo.