Self-hosted · Open Source · MIT License

Your AI is running.
You have no idea
what it's doing.

The self-hosted cockpit for Claude Code, Codex CLI & Gemini CLI sessions. See every session, every cost, every output — in real time. On your server.

Runs on your server. No one else sees your data. Free forever.
Install via Docker ↓ View on GitHub ↗
session-pilot — bash
# 3 Befehle. ~2 Minuten. Läuft.
git clone https://github.com/web-werkstatt/session-pilot.git
cd session-pilot && cp .env.example .env
docker compose up -d

✓ Running at http://localhost:5055
✓ Sessions synced: 341 sessions across 5 accounts
✓ Containers: 19 running · 0 unhealthy
341+ Sessions tracked
€240+ API costs analyzed
5 Accounts connected
€0 Subscription cost
The problem

You're not working with AI.
You're managing chaos.

Multiple sessions. Costs in the background. No history. No control.

  • PROBLEM Sessions run unattended. You have no idea what's happening right now, in which project, at what cost.
  • PROBLEM Costs are invisible. The API bill arrives. You have no breakdown. Nothing is attributable.
  • PROBLEM Outputs are gone. Claude did something. But what exactly? The session is closed. Nothing is findable.
  • PROBLEM No cross-project overview. Five projects, three accounts, zero visibility about which model ran where for how long.
Before SessionPilot — Status unknown
contypio/schema-editor running €??
seo-suite/collector unknown —
claude-session-4821 lost ???
web-werkstatt/redesign orphaned €12.40?
claude-session-3917 lost —
steuerrio/ea-tabelle unknown —
Total cost this month: ??? · Sessions reviewed: 0/47
Session Analysis Dashboard
The solution

You see every session.
Every cost.
Everything that actually happened.

No more guessing.

01
Sessions visible
See immediately what's running. Per project, per account, with full session history.
02
Costs visible
Token counts, model usage, per-project breakdown. No more surprises at the end of the month.
03
History retrievable
Every session logged. Full-text search. Rate sessions. Track what was actually done.
04
Your data stays with you
Self-hosted. No cloud. No telemetry. No one sees what your AI is doing except you.
Core Features

What you actually see.

Each feature has a clear purpose: make something visible that was invisible before.

Live
See what is running. Right now.

See immediately what Claude is doing. Full Markdown rendering, syntax highlighting, timestamps per message, navigable table of contents.

Costs
No more surprise invoices.

API costs broken down by model, project and timeframe. Token tracking, heatmaps, trend charts.

History
Know what Claude actually did.

Reconstruct, not guess. Every session logged, searchable, rateable. OK / Needs Fix / Reverted / Partial.

Infrastructure
Your containers — all healthy or not.

See which containers are running, healthy, or problematic. Live status, health checks, ports, uptime.

Time
Time tracked. No extra effort.

Automatic time tracking derived from session data. Know exactly how many hours each project consumed. Export-ready.

Quality
Code Quality Checks

7 automated checks: file sizes, duplication, cyclomatic complexity, CSS quality, architecture rules. Baseline and diff workflow.

Projects
Project Overview

Automatically detects projects, monorepos, and sub-projects. Tabbed detail view: sessions, plans, documents, relations.

Search
Full-Text Search

Ripgrep-powered search across all projects. Ctrl+K command palette for instant navigation.

Accounts
Multi-Account & Multi-LLM

Auto-discovers all AI coding accounts on your system. Claude Code, OpenAI Codex CLI, Google Gemini CLI — all in one dashboard.

Live
Usage Monitor

Real-time burn rates, plan limits (Pro/Max 5x/Max 20x), cost predictions, and reset timers — per account. Know exactly when you'll hit your limit.

At a Glance

The dashboard in action

Screenshots from real usage. What you see is what you get.

01 —
Session Overview
Session Overview

All sessions across all accounts. Filter by project, date, model.

02 —
Cost Analysis
Cost Analysis

API costs broken down by model, project, and period. No surprises.

03 —
Session Viewer
Session Detail View

Full Markdown rendering, code blocks with syntax highlighting, navigable TOC.

04 —
Docker Monitor
Docker Monitor

Live container status, health checks, port overview. Unhealthy containers notify immediately.

Quick Start

Running in 3 commands

No build step. Docker or interactive setup wizard. Linux, macOS, Windows. PostgreSQL optional — everything works without it except Sessions.

01
Clone & configure
git clone https://github.com/web-werkstatt/session-pilot.git && cd session-pilot && cp .env.example .env
02
Configure & launch
cp .env.example .env && docker compose up -d
03
Open
http://localhost:5055  # sessions auto-discovered

Or use the interactive setup wizard: ./setup.sh (Linux/macOS) or setup.ps1 (Windows) — checks prerequisites, finds free ports, sets up the database.

Pricing
€0
Free forever · No limits · No tricks
  • ✓Full source code on GitHub
  • ✓MIT License — commercial use permitted
  • ✓No API key required
  • ✓No cloud, no telemetry, no SaaS
  • ✓Works offline on your own server
  • ✓Docker & systemd deployment
View on GitHub ↗
Ownership

Your server.
Your data.
Your rules.

No SaaS dependency
SessionPilot runs entirely on your own infrastructure. No external service that can go down, change pricing, or lock you in.
No telemetry
Nothing is transmitted to third parties. Your session data, your project structure, your costs — all stays local.
Stack you know
Python + Flask + PostgreSQL. No magic, no black box. You can read, modify and extend every line yourself.
Tech Stack

What's inside

No magic dependencies. No vendor lock-in. Standard tooling that runs anywhere.

  • Python 3.9+ Application REQUIRED
  • Flask Web Framework REQUIRED
  • PostgreSQL 14+ Session Storage OPTIONAL
  • Docker Container Monitoring OPTIONAL
  • ripgrep Full-text Search OPTIONAL
  • Gitea Git Integration OPTIONAL
Who is this for

Built for developers
who actually use Claude Code.

Solo Developer
You run multiple Claude Code sessions across multiple projects.
SessionPilot gives you a single place to track costs, review what happened, and stay in control — without switching between terminals and log files.
Small Agency / Team
Multiple developers, multiple accounts, one shared server.
Track AI usage per project, attribute costs to clients, and keep your infrastructure visible — all self-hosted, no SaaS dependency.
FAQ

Frequently asked questions

SessionPilot supports Claude Code (full cost tracking, reviews, timesheets), OpenAI Codex CLI (session import and browsing), and Google Gemini CLI (session import and browsing). All accounts are auto-discovered — no manual configuration needed.
No. PostgreSQL is optional. Everything works without it except the Sessions feature, which requires a database to persist session data. Docker, ripgrep, and Gitea are optional too.
No. SessionPilot runs entirely on your own infrastructure. No telemetry, no analytics, no external API calls. Your session data, costs, and project structure stay on your server.
Yes — all three platforms are fully supported. Docker works everywhere. For native setup, run ./setup.sh on Linux/macOS or setup.ps1 on Windows. The interactive wizard checks prerequisites, finds free ports, and sets up the database automatically.
There is no catch. SessionPilot is MIT-licensed open source software. You self-host it, you own it. No freemium, no usage limits, no paid tiers — ever.
Pull the latest version from GitHub and restart: git pull && docker compose up -d --build. That's it.
Get started

You've been guessing
long enough.
Start seeing.

3 commands. 2 minutes. Everything visible — on your own server.

Install via Docker ↓ View on GitHub ↗

Free forever · Self-hosted · MIT License · No registration