Claudio - Navegador de Sessões Claude Code Local
shahfazal/claudio
Navegue sessões Claude Code localmente com interface web e análise de custos
Ver no GitHubPlugin que permite navegar e analisar suas sessões do Claude Code localmente, sem cloud ou contas. Lê o diretório ~/.claude/ e oferece uma interface web para explorar transcripts completos, histórico de comandos, estimativas de custo e memória de projetos.
Instalação
npx convez-ai shahfazal/claudio1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Auditar e revisar transcripts de sessões de código passadas
- Estimar custos de tokens por sessão e projeto
- Acessar memória de projetos e histórico de comandos localmente
- Pesquisar e organizar sessões por projeto sem sincronização
Veredito

Ferramenta bem pensada para quem usa Claude Code regularmente. Oferece visibilidade local sobre sessões e custos sem dependências externas, com interface clara e funcionalidades úteis (busca, temas, histórico). Requer uv e Python, mas setup é trivial.
README
A tradução para o português deste README ainda está na fila.
claudio
Browse your Claude Code sessions in a local web app.
Reads ~/.claude/ — the same directory Claude Code writes to. No cloud, no sync, no accounts.
What you get
- All sessions grouped by project, sorted newest-first
- Session titles from compact summaries ("Primary Request and Intent") or the first user message, no API calls, fully local
- Full transcript per session: user turns, assistant responses, tool calls
- Per-session command history and todo sidebar
- Estimated Costº
- Live search across titles and project paths
- Estimated cost per session (see note below)
- Per-project memory browser: reads Claude Code's
memory/files, linked from the index and session sidebar - Dark / light / system theme (persisted in
localStorage) - Click-to-copy for the project path
º - read below for Cost Estimates
Requirements
- uv: handles Python and dependencies automatically
- Claude Code installed (so
~/.claude/projects/exists)
Install & run
git clone https://github.com/shahfazal/claudio
cd claudio
uv run claudio
Open http://127.0.0.1:5001. That's it — uv creates a venv and installs dependencies on first run.
Environment variables
| Variable | Default | Description |
|---|---|---|
PORT | 5001 | Port to listen on |
HOST | 127.0.0.1 | Bind address |
DEBUG | (off) | Set to 1 for Flask debug mode |
Development
uv run python -m pytest # tests across parsers, routes, templates, health, retention, export, pricing
uv run python -m ruff check . # lint
uv run python -m ruff format . # format
Project layout
src/claudio/
├── app.py Flask app + routes
├── health.py Environment health checks
├── parsers.py JSONL parsing, data loading
├── pricing.default.json Bundled pricing rates (copied to ~/.claudio/ on first run)
└── templates.py Jinja2 template strings
tests/
├── fixtures/ Sample JSONL for tests
├── conftest.py
├── test_export.py
├── test_health.py
├── test_parsers.py
├── test_pricing.py
├── test_routes.py
└── test_templates.py
Example session path
~/.claude/projects/-Users-aturing-claude-worktrees-enigma-decrypt-happy-colossus/7ad5ba28-b562-4ce3-9d76-13c02ec7c4da.jsonl
Cost estimates
The cost shown per session is a locally calculated estimate, not a figure from Anthropic. Claudio reads the token usage logged in each assistant message (input_tokens, output_tokens, cache tokens) and multiplies by the published per-model rates.
A few caveats:
- Anthropic does not write a cost field to
~/.claude/session files, so there is no authoritative number to read. - Rates are loaded from
~/.claudio/pricing.json(editable). Updatelast_updatedand the relevant model rates when Anthropic changes pricing. See/healthin the app for a staleness warning after 90 days. - Sessions that used a model not in the pricing table show
≈ ?rather than a potentially wrong number.
Close enough to be useful for understanding relative session cost, but not a billing figure.
Historical data depth
Claudio reads your session transcripts from ~/.claude/projects/. Those files
are subject to Claude Code's own retention sweep, controlled by
cleanupPeriodDays in ~/.claude/settings.json (default: 30 days).
Starting with Claude Code v2.1.117 (shipped right after v2.1.116 on April 20,
2026), this sweep runs on startup: session files older than cleanupPeriodDays
are deleted every time claude launches. Earlier versions largely let session
files accumulate, so if you used Claude Code before that release you may have
seen deeper history than you do now.
What this means for Claudio: with the default config, your stats only reach back 30 days. Anything older has already been removed from disk by Claude Code, and Claudio cannot recover what is no longer there.
To keep a longer window, raise the retention period before your next launch:
// ~/.claude/settings.json
{ "cleanupPeriodDays": 3650 }
Do not set it to 0. The docs imply 0 disables cleanup, but it currently
disables transcript persistence entirely, so you lose everything.
For analytics that survive the sweep regardless of this setting, export regularly with Claudio's JSON export (the Export button in the nav bar). Snapshots persist even after the underlying session files are gone.
Contributing
Contributions welcome! Before opening a PR:
- Run
uv run python -m ruff check .anduv run python -m ruff format . - Add tests for new features (see
tests/for examples) - Fill out the PR template with description and test plan
Legal: Contributions must be your own work, created on your own time and resources, not subject to employer IP claims. By submitting a PR, you confirm you have the right to license this work under MIT.
See CONTRIBUTING.md for full details.
⚠️ Stability & Compatibility
Claudio reads undocumented local files created by Claude Code:
~/.claude/projects/<slug>/<uuid>.jsonl: session data~/.claude/history.jsonl: session index~/.claude/projects/<slug>/memory/: memory files
This is fragile by design. These files, their locations, and their schemas are not a public API. If Anthropic changes where or how sessions are stored, Claudio will break.
What v0.4 does about it
| Feature | What it does |
|---|---|
| Health check | Validates ~/.claude/ structure and spot-checks session schema on every page load. Visit /health for details. |
| Export | Download all sessions as portable JSON via the nav bar. If Claudio breaks, you keep your data. |
| External pricing | Rates live in ~/.claudio/pricing.json, edit manually when Anthropic changes pricing, no code change needed. |
If Claudio breaks after a Claude Code update
- Check
/healthfor diagnostics - Export your data before investigating
- File an issue: https://github.com/shahfazal/claudio/issues
This is a hobby project, not production software. Enjoy it while it works.
Changelog
| Version | Commit | Feature |
|---|---|---|
| v0.7.0 | --- | Durable session store: live→store mirror, archive-backed stats & memory, calibrated retention detector |
| v0.6.1 | --- | Retention-sweep detector (detection-only; durable archival lands next minor) |
| v0.6.0 | d5e750f | Driver.js help tour; package version read at runtime |
| v0.5.0 | addb62c | Stats dashboard: heatmap, filter panel, cumulative cost chart |
| v0.4.0 | ac3917c | Resilience: health checks, export, external pricing |
| v0.3.0 | b9a60e0 | Compaction count |
| v0.2.0 | c7ef66d | Memory browser |
| v0.1.0 | b2bd054 | Initial release |
Disclaimer
The name "Claudio" is a nod to Claude Code sessions, and also just a guy's name. Any resemblance to AI assistants, living or dead, is purely coincidental.
Claudio is an independent open-source tool for browsing locally stored Claude Code session files on your own machine. It is not affiliated with, endorsed by, or sponsored by Anthropic. "Claude" is a trademark of Anthropic. Claudio simply reads files that Claude Code writes to ~/.claude/ and has no connection to Anthropic's servers or services beyond what Claude Code itself already does.
License
MIT