AI coding is
everywhere.
Clarity isn’t.
Claude Code, Copilot, Codex—AI activity is spread across tools and teams. See adoption clearly, spot opportunities for better AI practices, and find underused subscriptions. Turn scattered signals into smarter engineering decisions.
Free 30-day pilot on your own instance
- Claude Code
- Codex CLI
- GitHub Copilot
- Gemini CLI
- Antigravity
- OpenCode
Three questions your AI budget can finally answer.
Vendor dashboards show one tool each. StackSignal starts where the work happens: the sessions themselves.
Who is really using AI, in which tools?
Adoption by person, team and project, from the logs each AI tool keeps on the computer.
- The tool someone opened and the model that answered are counted apart, so Copilot running Claude is never counted twice.
- Quiet people and silent computers are told apart, so a broken agent never looks like a lazy engineer.
Is anyone using it well?
One AI-practice score from 0 to 100 for each person, team and project, with the evidence behind every point.
- The Briefing suggests who could use a coaching conversation, and how to open it: evidence for a conversation, not a verdict.
- It describes habits. It is not a performance rating, and there is no leaderboard.
Put every AI subscription to work.
LimitLens, part of StackSignal, answers four questions about every AI subscription: Claude Pro, Max and Team accounts, plus Codex, Copilot and Google seats.
- Who is blocked? Lockouts, people waiting, and when each account frees up.
- Who is close to the limit? 5-hour and 7-day windows, and the days spent at or near the limit.
- Where is capacity unused? Idle seats, and plans bigger than the work needs.
- Which plan should change? Nightly suggestions to upgrade, step down or reclaim a seat, each with its evidence and monthly cost change. A person decides; nothing changes an account by itself.
A coach for every developer, not a camera.
Each developer gets their own dashboard: their score, the changes worth the most points, and requests they repeat often enough to turn into a reusable skill.
- Tips come from 45 practice rules a developer can read.
- Five rules about working hours and tone never reach a manager or the score.
- The free StackSignal Coach for VS Code shows the same habits inside the editor.
"Your manager and the company's owners and administrators see this same page about you. It never shows a whole conversation." Shown on every developer's dashboard.
Live in a day. Nothing new for developers to do.
You get a dedicated instance: your data never shares a database with another company.
- 1
Install the agent
One small program per computer, on Windows, macOS or Linux. Developers enrol with their work email, or IT pushes it.
- 2
It reads and summarises
Every 15 minutes it reads the AI tools' local logs, read-only, and sends a summary over HTTPS. Updates are signed and checked before they run.
- 3
Everyone sees their own view
Leaders see the company, managers the team they lead, developers themselves. Access is enforced on the server.
What it collects, stated plainly.
Trust is the product. Here is what leaves a laptop, what never does, and who can read it. Full detail on the security and privacy page.
Run the migration with DB_PASSWORD=Xk29!pqLm and email the report to [email protected]
Run the migration with DB_PASSWORD=Xk29!pqLm and email the report to [email protected]
Sent every 15 minutes
- Counts and timings of AI sessions
- Tool, model and token use
- Practice findings and scores
- Short excerpts of requests, if your organisation keeps them on: most at most 300 characters, a few selected prompts up to 2,000
Never sent
- Whole conversations
- Files, attachments or images
- Commit messages
- A cost per person
Who reads what
- Developers: themselves
- Managers: the team they lead
- Owners and administrators: everyone
- Owners switch excerpts on or off for each computer
Works with the tools your team already uses.
Read from each tool's own local logs. Nothing to install inside the tools, and nothing changes in how people work.
- Claude CodeTerminal and IDE sessions
- Codex CLIOpenAI Codex, terminal sessions
- GitHub CopilotVS Code, the CLI and the app
- Gemini CLITerminal sessions
- AntigravityEditor and CLI sessions
- OpenCodeTerminal sessions
Not yet supported: Cursor, Windsurf and JetBrains AI. Seat and quota readings: Claude (5-hour and 7-day windows), Codex, and a weekly Google reading; GitHub publishes no Copilot quota. Product names and logos are trademarks of their owners, shown only to describe compatibility; StackSignal is not affiliated with or endorsed by them.
Priced per active developer. Only people who use AI count.
A person counts for a month if their computer reported AI activity that month. You can check the number yourself in the console.
Coach
- The StackSignal Coach in VS Code
- Your own habits, score and tips
- Runs on your computer and sends nothing to StackSignal
Team
- Leader and manager console
- Practice score, Briefing, coaching
- AI seats and spend
- Agents for Windows, macOS, Linux
- Dedicated hosted instance
Business
- Everything in Team
- Microsoft sign-in
- APIs for AI-levels programmes
- Weekly emailed reports (beta)
- Priority support, quarterly review
Enterprise
- Self-hosted or in your cloud
- Data processing agreement and security review
- Client evidence packs for IT services firms
- Named success manager
Free 30-day pilot. Up to 25 developers on a dedicated instance, a setup call, and a findings read-out at the end. No card needed.
Start a 30-day pilotQuestions buyers ask first.
Anything else: support, or book a demo and ask us directly.
Is this employee monitoring?
It measures how people work with AI tools, not how much they work. It shows habits and seat use, never a productivity or performance rating, and there are no leaderboards. Five rules about working hours and tone are hidden from managers and the score. It does show when AI was used, and with excerpts on, short parts of requests. Tell your people what it collects before you roll it out; we give you the text.
Which AI tools does it read?
Claude Code, OpenAI Codex CLI, GitHub Copilot in VS Code, the Copilot CLI and app, Gemini CLI, Antigravity and OpenCode. Cursor, Windsurf and JetBrains AI are not read yet.
Can we keep prompt text out of it?
Yes. Excerpts can be switched off for any computer by an Owner. With excerpts off, counts, timings, scores and up to six topic keywords per repeated request are still sent; request text is not.
How is the AI-practice score calculated?
A weighted average of four parts for the days you pick: regular use (30%), depth of work (25%), prompt quality (25%) and cost efficiency in tokens per hour (20%). Missing data is shown as missing, never scored as zero.
Where does our data live?
On a dedicated instance for your company only. Enterprise customers can run it in their own cloud or on their own servers.
Does it change our AI accounts or plans?
No. The seats module observes and recommends. A person decides, and someone still changes the plan with the vendor.
When can developers get the free Coach?
The VS Code Coach is being prepared for the Visual Studio Marketplace and Open VSX. Book a demo to hear when it is live.
Do you have SOC 2 or ISO 27001?
Not yet. We share our security pack, data-flow diagram and answers to your questionnaire, and a penetration test is planned before general availability.
See your own team's AI practice in 30 days.
Install the agent on up to 25 computers. At the end of the month we walk you through what we found, and you decide.