pisama-claude-code

Lightweight trace capture for Claude Code sessions with token usage and cost tracking.

Demo

Why Pisama?

When working with Claude Code, have you ever wondered:

• How much did that session cost? Track token usage and costs in real-time
• What tools were called? See every Bash, Read, Write, and Edit operation
• Why did it fail? Capture traces for debugging and forensics
• Can I export my sessions? JSONL export for analysis or compliance

pisama-claude-code captures everything Claude Code does, locally and privately.

┌─────────────────────┐         ┌─────────────────────┐
│   Claude Code       │         │   Pisama Platform   │
│   + pisama-cc       │ ──────▶ │   (optional)        │
│   (capture)         │  sync   │   - detection       │
└─────────────────────┘         │   - self-healing    │
        │                       └─────────────────────┘
        │
        ▼
   ~/.claude/pisama/traces/
   (local storage)

Installation

pip install pisama-claude-code

Requirements: Python 3.10+ and Claude Code CLI

Quick Start

# 1. Install capture hooks
pisama-cc install

# 2. Use Claude Code normally - traces are captured automatically

# 3. View your session data
pisama-cc status        # Summary with token totals and cost
pisama-cc traces        # Recent tool calls
pisama-cc usage         # Detailed breakdown

Features

Token & Cost Tracking

$ pisama-cc usage --by-model --by-tool

📊 Token Usage Summary (last 100 traces)
==================================================
Input tokens:           10,234
Output tokens:          85,421
Cache read tokens:   1,234,567
Total cost:        $    52.34

📈 By Model:
--------------------------------------------------
  claude-opus-4-5-20251101            $52.34

🔧 By Tool:
--------------------------------------------------
  Bash                   45 calls  $25.12
  Read                   30 calls  $15.34
  Write                  20 calls  $8.45
  Edit                   5 calls   $3.43

Session Status

$ pisama-cc status

📊 Pisama Status
========================================

🔧 Hook Installation:
   ✅ pisama-capture.py
   ✅ pisama-post.sh
   ✅ pisama-forward.py
   ✅ pisama-forward.sh
   All hooks installed

📁 Local Traces: 1,400
   Input tokens:  9,580
   Output tokens: 79,569
   Total cost:    $43.22

Export & Analysis

# Export to JSONL
pisama-cc export -o traces.jsonl

# Export compressed
pisama-cc export -o traces.jsonl.gz --compress

# Export to OpenTelemetry format
pisama-cc export --format otel -o traces-otel.json

# Filter by date range
pisama-cc traces --since 2025-01-01 --until 2025-01-04

OpenTelemetry Integration

Export traces to any OTEL-compatible backend (Jaeger, Honeycomb, Datadog, etc.):

# Install OTEL support
pip install pisama-claude-code[otel]

# Export to local Jaeger
pisama-cc export-otel -e http://localhost:4318/v1/traces

# Export to Honeycomb
pisama-cc export-otel -e https://api.honeycomb.io/v1/traces \
    -H "x-honeycomb-team=YOUR_API_KEY"

# Export to file in OTEL format
pisama-cc export --format otel -o traces.json

OTEL export uses GenAI semantic conventions for token usage, costs, and model attributes.

CLI Reference

Command	Description
pisama-cc install	Install capture hooks to ~/.claude/hooks/
pisama-cc uninstall	Remove hooks
pisama-cc status	Show status, token totals, and cost
pisama-cc traces	View recent traces (-v for verbose, -c for content)
pisama-cc usage	Token usage breakdown (--by-model, --by-tool)
pisama-cc export	Export to JSONL or OTEL (--format otel, --compress)
pisama-cc export-otel	Export to OpenTelemetry collector (-e ENDPOINT)
pisama-cc connect	Connect to Pisama platform (forwarding is opt-in; add --auto-sync to forward every session)
pisama-cc sync	Upload traces to platform
pisama-cc analyze	Run failure detection (requires platform)
pisama-cc proxy serve	Run the opt-in reasoning proxy for a session (experimental)
pisama-cc proxy install --always-on	Always-on reasoning capture (macOS launchd)
pisama-cc proxy status / uninstall	Proxy health / teardown
pisama-cc vault status	Show PII tokenization vault status ([core])

Cost Estimates

Cost estimates use Anthropic's first-party API list prices per 1M tokens. The current active model families are:

Model family	Input	Output	Cache write	Cache read
Claude Opus 4.5 to 4.8	$5.00	$25.00	$6.25	$0.50
Claude Sonnet 5	$2.00	$10.00	$2.50	$0.20
Claude Sonnet 4.5 to 4.6	$3.00	$15.00	$3.75	$0.30
Claude Haiku 4.5	$1.00	$5.00	$1.25	$0.10

Sonnet 5 introductory pricing is applied through August 31, 2026, then automatically changes to $3 input, $15 output, $3.75 cache write, and $0.30 cache read. Provider-specific discounts, batch pricing, long-context premiums, and regional pricing are not included. Check Anthropic pricing for the authoritative rates.

Privacy & Security

• Local-first: all traces are stored in ~/.claude/pisama/traces/.
• Private on disk: trace, proxy, config, sync-log, and lite-mode files use user-only permissions on POSIX systems. Configuration writes are atomic.
• No hidden raw copy: captured hook fields are stored once. Sensitive tool input is no longer duplicated in an untokenized raw payload.
• Forwarding is opt-in: connect defaults to no auto-sync. Nothing leaves your machine until you run pisama-cc sync or reconnect with --auto-sync.
• Secrets scrubbed by default: credential-shaped strings (API keys, JWTs, cloud tokens) are redacted from content before forwarding — no extras needed. The optional [core] extra adds pisama-core's reversible keychain vault.
• Paths anonymized: home-directory paths are replaced with ~.
• Reasoning proxy is experimental + opt-in: pisama-cc proxy routes API traffic through a local logging proxy to recover extended-thinking. It defaults to API-key usage. Subscription (OAuth) users: it handles your account token, which is ToS-sensitive — use only knowingly.

See SECURITY.md for our security policy.

Configuration

After installation, the hooks are automatically configured. To customize, edit ~/.claude/settings.local.json:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          { "type": "command", "command": "~/.claude/hooks/pisama-post.sh", "timeout": 10, "async": true }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          { "type": "command", "command": "~/.claude/hooks/pisama-forward.sh", "timeout": 30, "async": true }
        ]
      }
    ]
  }
}

Platform Integration (Optional)

For advanced features like failure detection and self-healing, connect to the Pisama platform:

pisama-cc connect --api-key <key>  # Authenticate
pisama-cc sync                     # Upload traces
pisama-cc analyze                  # Run detection

Platform features:

• 25 MAST failure mode detection
• AI-powered fix suggestions
• Self-healing automation
• Visual dashboard

Part of the Pisama Platform

pisama-claude-code is the Claude Code integration for the broader Pisama multi-agent failure detection platform, which supports multiple agent frameworks:

Framework	Package	Status
Claude Code	pisama-claude-code	Stable
LangChain/LangGraph	pisama-core SDK	Available
CrewAI	pisama-core SDK	Available
AutoGen	pisama-core SDK	Available
n8n	pisama-core SDK	Available

For other frameworks, see the Pisama integration docs.

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

# Development setup
git clone https://github.com/Pisama-AI/pisama-claude-code.git
cd pisama-claude-code
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

# Run tests
pytest --cov=pisama_claude_code --cov-fail-under=60

# Run the same quality gates as CI
ruff check src/ tests/
ruff format --check src/ tests/
mypy src/pisama_claude_code/
xenon --max-absolute D --max-modules C --max-average B src/pisama_claude_code
pylint --disable=all --enable=duplicate-code --min-similarity-lines=12 src/pisama_claude_code

Changelog

See CHANGELOG.md for release history.

License

MIT License - see LICENSE for details.

Links

• Documentation
• Pisama Platform
• Issue Tracker
• Discussions

---

Made with ❤️ for the Claude Code community