You're doing more than ever. You need mission control to keep it coherent.

Every AI tool manages the agent's context. Nobody manages yours. HumanContext fixes that.

The Problem

You're running 3-5 projects across Claude Code, Cursor, and Windsurf. Each tool knows about its own session but has no idea what you did in the others. Every morning, you spend 30+ minutes re-orienting: which branch was I on? What did the agent last do? What decisions are pending? How much am I spending?

The Solution

$ npx humancontext

Your browser opens. You see everything:

• Project cards — git state, latest session summary, activity breakdown per project
• Trend charts — how your coding patterns change week over week
• Cost tracking — what you're spending across AI tools, projected monthly
• Auto-reports — daily summary when your coding day ends, weekly digest every Monday
• Decision quality — which AI suggestions you accept vs reject, and why

No config. No accounts. No cloud. Everything runs on your machine.

Quick Start

# One command — auto-detects your project and LLM, opens dashboard
npx humancontext

# Or install globally
npm install -g humancontext

# Dashboard opens automatically on first run
# After that:
hc dashboard        # Open the dashboard
hc briefing         # Morning briefing in terminal
hc insights         # Analytics in terminal
hc insights --cost  # Cost report
hc insights --weekly # Weekly digest

Dispatch — Control from Your Phone

# Enable remote control (LAN access)
hc dispatch

A QR code appears. Scan it with your phone. You're now controlling your dev machine's HumanContext from your phone — generate briefings, force reports, check project status. All on the same WiFi. No cloud, no accounts.

Connect Slack (30 seconds)

Get daily reports and weekly digests pushed to your Slack channel:

# 1. Create an incoming webhook in Slack (Settings → Apps → Incoming Webhooks)
# 2. Copy the webhook URL
# 3. One command:
hc config slack_webhook https://hooks.slack.com/services/T.../B.../xxx

That's it. Reports will push to Slack automatically.

How It Works

1. You code. HumanContext watches your AI sessions in the background.
2. You stop. After 30 minutes of inactivity, it generates a daily report.
3. You get notified. OS notification: "Your daily report is ready." Click → dashboard opens.
4. Monday morning. Weekly digest auto-generates. Dashboard shows this week vs last week. Slack gets the same digest.

Dashboard

The interactive dashboard at localhost:7272 is the primary way to use HumanContext:

• Project cards with git state, latest session, activity distribution
• Trend charts showing activity, cost, and cognitive patterns over time
• Auto-generated reports — daily summaries and weekly digests front and center
• Cost intelligence — spending by project, tool, and week with projected monthly
• Decision quality — AI acceptance/rejection rates with rationale breakdowns

Auto-refreshes when new data arrives. No login. No cloud. Your browser, your data.

All Commands

| Command | Description | |---------|-------------| | npx humancontext | Auto-detect project + LLM, open dashboard | | hc dashboard | Open the interactive dashboard | | hc init [path] | Register a project directory | | hc status | Quick overview of all projects | | hc briefing | Morning briefing across all projects | | hc briefing --raw | Data-only briefing, no LLM needed | | hc insights | Analytics snapshot | | hc insights --cost | Cost breakdown | | hc insights --weekly | Weekly digest | | hc projects | List tracked projects | | hc remove <id> | Stop tracking a project | | hc config <key> [value] | Get or set configuration | | hc dispatch | Remote control from phone (QR code pairing) | | hc providers | Show available LLM providers | | hc api | Start the HTTP API server |

Intent Engine

HumanContext includes the @humancontext/intent-engine for deep analysis of your AI coding sessions. Every session runs through a 4-layer pipeline:

Layer 1 — Activity Classification classifies every event into 7 categories: navigation, editing, debugging, testing, refactoring, comprehension, other. Time-split percentages included.

Layer 2 — Cognitive Phases detects workflow stages (understanding, hypothesizing, experimenting, verifying) with phase hysteresis. Tracks iteration cycles and phase transitions.

Layer 3 — Intent Narrative produces a human-readable summary: what you were doing, which files were involved, what approaches were tried. Includes 10 debugging dimensions.

Layer 4 — Decision Capture identifies every point where you accepted, rejected, or redirected an AI suggestion, with rationale categorization.

Pro Analytics (free for 15 active coding days)

All analytics are derived from data already collected by the intent engine — no new tracking.

| Category | What You Learn | |----------|---------------| | Activity Trends | How your editing vs debugging vs navigation time shifts over weeks | | Cognitive Flow | Phase patterns, iteration cycles, thrashing detection | | AI Effectiveness | Rejection rates by tool, confidence trends | | Cost Intelligence | Spending by project/tool/week, projected monthly | | Debugging Intelligence | Bug severity, root cause, and reproducibility patterns | | Productivity Patterns | Peak hours, active days, session duration trends | | Weekly Digest | This week vs last week — what changed, what to watch |

An "active coding day" = a day where HumanContext detects at least one AI session. Weekend coders get ~5 weeks of trial. Free features never stop working.

Session Discovery

Auto-discovers AI sessions from multiple tools using 4-signal project matching:

| Format | Source Tools | |--------|-------------| | Claude Code JSONL | Claude Code | | CASS JSON/JSONL | 13+ agents (Claude Code, Cursor, Windsurf, Aider, etc.) | | Agentlytics JSON | 16+ editors (Cursor, Windsurf, Copilot, etc.) | | Agentlytics SQLite | Same as above (local cache) |

LLM Intelligence Chain

Auto-detects the best available AI provider. No manual configuration required.

| Priority | Provider | Privacy | Setup | |----------|----------|---------|-------| | 1 | Ollama (local) | Nothing leaves your machine | Install Ollama | | 2 | LM Studio (local) | Nothing leaves your machine | Install LM Studio | | 3 | Your API key | Your key, your data policy | Set ANTHROPIC_API_KEY, OPENAI_API_KEY, etc. | | 4 | No AI | N/A | Always available — raw data only |

Supports 11 providers: Ollama, LM Studio, LocalAI, vLLM, Anthropic, OpenAI, Groq, OpenRouter, Mistral, Together, Gemini.

MCP Server (9 tools + 2 resources)

Add to your Claude Desktop or Cursor config:

{
  "mcpServers": {
    "humancontext": {
      "command": "npx",
      "args": ["-y", "humancontext", "mcp"]
    }
  }
}

Then ask your AI assistant: "What's happening across all my projects?" or "Show me my cost trends."

MCP Tools

| Tool | Tier | Description | |------|------|-------------| | get_project_status | Free | Current state of all projects | | get_morning_briefing | Free | Full morning briefing | | get_project_decisions | Free | Recent decisions and AI session data | | analyze_session | Free | Run 4-layer analysis on a session file | | get_context | Free | Auto-discover latest session and summarize | | list_sessions | Free | Find AI sessions on disk | | get_analytics | Pro | Full analytics dashboard data | | get_weekly_digest | Pro | Weekly comparison across projects | | get_cost_report | Pro | Cost breakdown and projections |

Report Delivery

Reports are generated automatically and delivered to all connected channels:

| Channel | Setup | What You Get | |---------|-------|-------------| | Dashboard | Automatic | Daily/weekly reports rendered with charts | | Slack | hc config slack_webhook <URL> | Formatted blocks in your channel | | OS Notification | Automatic | "Report ready" → click opens dashboard | | CLI | hc insights | On-demand in terminal | | MCP | Ask your AI assistant | Analytics via tool calls |

Architecture

Layer 1 — WATCH      File watcher, Git tracker, AI session reader
                     (auto-discovers Claude Code, CASS, Agentlytics sessions)

Layer 2 — STRUCTURE  SQLite database (WAL mode, < 50MB)
                     Projects, git state, file activity, AI sessions,
                     briefings, config, analytics data

Layer 3 — THINK      Intent engine (4-layer analysis per session)
                     + Briefing engine + LLM intelligence chain
                     + Analytics engine (7 Pro categories)
                     + Auto-report scheduler

Layer 4 — SURFACE    Dashboard (localhost:7272) ← PRIMARY
                     + CLI (11 commands) + MCP server (9 tools)
                     + HTTP API + Slack webhook + OS notifications

Everything runs on your machine. The database is at ~/.humancontext/state.db. No cloud. No telemetry. No accounts.

Privacy

HumanContext is privacy-first by architecture, not by policy. The database is local SQLite. The default LLM is local Ollama. There is no server, no account, no telemetry. Your code never leaves your machine.

Testing

226 tests across 2 packages, 0 failures:

npm test                    # Run all HC CLI tests (83)
cd ../hc-intent-engine && npm test  # Run intent engine tests (143)

Contributing

See CONTRIBUTING.md. All contributors must sign a CLA before their first PR is merged.

License

Apache 2.0 — see LICENSE.