DEEP_DIVE

How UseAI Works

What gets captured, how it's measured, and what it means for your career. This is the complete guide to every metric, score, and signal UseAI produces.

ACT_01

What UseAI Captures

Every Prompt, Captured Automatically

PROMPT_LIFECYCLE

Every time you work with an AI tool, UseAI silently records the full prompt lifecycle — from the first message to the final evaluation. No manual logging, no forms to fill out, no context switching. The background daemon captures everything.

01

Start

useai_start

Session begins when your AI tool sends the first message. Tool, task type, and project are recorded automatically.

02

Track

useai_heartbeat

Heartbeats keep the session alive during long prompts, tracking active time intervals for accurate clock time calculation.

03

Seal

useai_end

Session closes with a full evaluation, Ed25519 signature, and hash chain entry. Any change from this point forward breaks the signature.

Not All Output Is Equal

OUTPUT_BREAKDOWN

Features shipped. Bugs fixed. Refactors completed. Tests written. Every milestone you complete is categorized by type and weighted by complexity — because a complex architecture overhaul is not the same as a quick typo fix.

Complexity weights: simple ×1 · medium ×2 · complex ×4. Your output fingerprint shows what kind of developer you really are.

Feature

New functionality shipped

Bug Fix

Defect identified and resolved

Refactor

Structural improvement, same behavior

Test

Test coverage added or improved

Docs

Documentation written or updated

Setup

Project scaffolding or tooling

Deployment

Released to production

Other

Miscellaneous development work

Where Your AI Hours Go

TIME_INTELLIGENCE

Are you mostly debugging or mostly building? UseAI breaks down your AI time by task type, giving you real numbers for how AI fits into your workflow — daily, weekly, and monthly.

Active session time is measured via heartbeats, not wall clock. If you step away for coffee, that gap isn't counted. You see real time spent with AI, not calendar time.

coding

Building new features and writing implementation code

debugging

Investigating and fixing bugs, tracing error paths

testing

Writing and running tests, verifying behavior

planning

Architecture decisions, task breakdown, scoping

reviewing

Code review, PR feedback, quality checks

documenting

Writing docs, READMEs, inline comments

learning

Exploring new tools, libraries, or concepts

ACT_02

How UseAI Measures

AI Proficiency Evaluation

AI_PROFICIENCY

At the end of every prompt, the AI evaluates how effectively you used it across four dimensions. Each dimension is scored 1–5, and for any score below 5, the AI provides a concrete, actionable reason explaining what was missing and how to improve next time.

These scores feed into your APS and public profile — they're not just feedback, they're the foundation of your AI proficiency signal.

PROMPT QUALITY

How clear, specific, and well-structured your prompts are. High scores mean the AI rarely needs clarification and can act on your instructions immediately.

CONTEXT PROVIDED

How much relevant context you give the AI upfront — file references, constraints, background. Better context means fewer round-trips and more accurate output.

SCOPE QUALITY

How well-scoped your tasks are. A focused, single-responsibility prompt scores higher than a vague or overly broad request that tries to do too much at once.

INDEPENDENCE LEVEL

How autonomously the AI can work from your instructions. High independence means you gave enough direction for the AI to execute without constant hand-holding.

AI Proficiency Score (APS)

AI_PROFICIENCY_SCORE

The APS is a composite 0–1000 score that aggregates your performance across multiple sessions. It combines five components, each normalized to 0–1 and weighted to produce a holistic measure of AI-assisted development proficiency.

APS captures your entire body of work — rewarding consistency, breadth of skills, and sustained output over time.

Output

25%

Complexity-weighted milestones completed in the window

Efficiency

25%

Complexity weight per hour of AI session time

Prompt Quality

20%

Average prompt quality, context, and scope scores

Consistency

15%

Active days ratio, streak, and session frequency

Breadth

15%

Unique languages, AI tools, and tool leverage score

ACT_03

What It Means For You

Your AI Developer Identity

DEVELOPER_IDENTITY

GitHub shows your commits. UseAI shows what you built with AI and how effectively you wield it. A public, shareable profile displaying your tools, languages, output volume, complexity distribution, and proficiency scores — your AI development resume.

In a world where every developer “uses AI,” prove you don't just use it — you're proficient with it.

PUBLIC PROFILE

A shareable page showing your AI activity — tools, languages, output volume, complexity, and proficiency scores. Only aggregate stats are shown — no titles, project names, or code.

LEADERBOARD RANKING

See where you stand globally. APS ranks developers by output, efficiency, prompt quality, consistency, and breadth.

PROFESSIONAL SIGNAL

Visible to recruiters, teams, and the community. Demonstrate AI proficiency with cryptographically sealed session data.

Cryptographically Sealed

VERIFICATION

Every prompt is sealed with an Ed25519 signature and linked in a hash chain. A real-time seal verification call to the cloud proves the session happened when it claims — only verified sessions count towards the leaderboard.

SEALED SESSIONS

Every completed session is sealed with an Ed25519 digital signature and includes milestones in the signed data. Any edit after sealing invalidates the signature and is detected on read or upload.

HASH CHAIN INTEGRITY

Sessions are linked in a SHA-256 hash chain. Each entry references the previous one. Tampering with any record breaks the chain and is immediately detectable.

SEAL VERIFICATION

At session end, a verification request is sent to the cloud with the session ID and timestamp. The server generates a unique signature — proving the session was sealed in real-time, not fabricated later.

Your Data, Your Machine

PRIVACY

No source code or prompt contents are ever transmitted. The daemon processes everything locally in ~/.useai. When you sync, session metadata, titles, project names, evaluation scores, and milestones are sent. You own your raw data — always.

The entire project is open source under the AGPL-3.0 license. You can audit every line of code that runs on your machine.

NO SOURCE CODE

No source code or prompt contents are ever transmitted. Session metadata, titles, project names, evaluation scores, and milestones are synced.

YOU CONTROL VISIBILITY

Private titles, project names, and evaluation reasons are synced but only visible to you. Public profiles show aggregate stats only — no titles, no project names, no code.

LOCAL PROCESSING

The UseAI daemon runs on your machine and stores data in ~/.useai. A seal verification call is made at session end for leaderboard eligibility — if offline, the session still seals normally.

DATA OWNERSHIP

Your session history is stored as date-based JSONL files (e.g. 2026-04-27.jsonl) you can read, export, or delete at any time. No vendor lock-in. Your data belongs to you.

REFERENCE

Supported Tools

Works With Your Stack

30 tools listed

Claude Code

Anthropic's CLI coding assistant

Claude Desktop

Anthropic desktop app

Cursor

AI-first code editor

Windsurf

AI-powered IDE

VS Code

Visual Studio Code MCP client

VS Code Insiders

Insiders channel for VS Code

Codex

OpenAI coding assistant

Gemini CLI

Google's AI coding CLI

GitHub Copilot

AI pair programmer by GitHub

Copilot CLI

GitHub Copilot terminal assistant

Aider

AI pair programming in terminal

Amazon Q CLI

Amazon Q command-line assistant

Amazon Q IDE

Amazon Q in IDE workflows

Zed

Zed editor integration

Cline

VS Code autonomous coding agent

Roo Code

Roo coding agent extension

Kilo Code

Kilo Code VS Code extension

Trae

Trae editor integration

Antigravity

Antigravity / Gemini integration

Goose

Goose desktop coding agent

OpenCode

OpenCode terminal coding agent

Crush

Crush coding assistant

Junie

JetBrains Junie assistant

JetBrains

JetBrains IDE integration

Continue

Continue coding assistant

Sourcegraph Cody

Sourcegraph AI coding assistant

TabbyML

Self-hosted code assistant

Augment

Augment coding assistant

Amp

Amp coding assistant

MCP Client

Generic MCP-compatible client

Want to add your tool?

UseAI works with any MCP-compatible AI tool.

View on GitHub