ShekharBhardwaj/AgenticLedger

Overview

AgenticLedger is a local-first flight recorder for AI agents. It captures every LLM call, tool call, cost, and loop through a transparent proxy, requiring zero code changes. The tool is written in Python and stores data locally.

Best for

Best for
Developers debugging AI agents locally without code changes

Use cases

• Debug agent loops by reviewing captured call sequences
• Track token and tool usage costs across sessions
• Audit agent behavior without modifying application code

How to use

Install

pip install -U agentic-ledger

Tools exposed

• action_id
• session_id
• model_id
• system_prompt
• tool_calls
• tool_results
• stop_reason
• cost_usd
• latency_ms
• status_code
• error_detail
• agent_name
• user_id
• app_id
• parent_action_id
• list_sessions
• list_runs
• AGENTICLEDGER_UPSTREAM_URL
• AGENTICLEDGER_DSN
• AGENTICLEDGER_HOST

Tested with

Claude Desktop, Claude Code, Cursor, Continue, ChatGPT

Notes

AgenticLedger is a local-first flight recorder for AI agents. It captures every LLM call, tool call, cost, and loop through a transparent proxy, requiring zero code changes. The tool is written in Python and stores data locally.

5 stars on GitHub. Last updated 2026-08-30. Licensed MIT.

Use cases

• Debug agent loops by reviewing captured call sequences
• Track token and tool usage costs across sessions
• Audit agent behavior without modifying application code

Pros

• Local-first design keeps sensitive data on the machine
• Transparent proxy means no code changes are needed
• Captures comprehensive data including LLM, tool, cost, and loop details

Cons

• Requires routing agent traffic through the proxy
• Local storage may not suit distributed or remote teams
• Limited community adoption (5 stars) suggests early-stage maturity

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.