Skip to main content
CodeFire provides real-time monitoring of active Claude Code sessions and maintains a searchable history of all past sessions with detailed analytics.

​
Live Session View

The Dashboard tab shows a live feed of your active Claude Code session:
Live session monitoring with cost tracking

​
Real-Time Metrics

CodeFire polls the session JSONL file every 2 seconds and displays:
  • Context usage — Current tokens vs. 200k limit with visual meter
  • Cost — Estimated spend based on model pricing
  • Messages — Total user + assistant messages
  • Tools — Number of tool invocations (Read, Edit, Bash, etc.)
  • Elapsed time — How long the session has been running
CodeFire reads only new bytes from the session file on each poll, making monitoring lightweight even for large sessions.

​
Tool Usage Breakdown

Expand the “Tool Usage & Files” section to see:
  • Bar chart of tool invocations (Read, Edit, Bash, Glob, etc.)
  • Files changed — List of files the agent has touched
  • Tool counts — How many times each tool was called

​
Activity Feed

The live feed shows a chronological stream of events:
  • 🔵 User messages — Your prompts to the agent
  • 🟢 Assistant text — Claude’s responses
  • 🟣 Tool use — Each tool invocation with context (file path, command, etc.)
Scroll through the feed to see what the agent is doing in real-time.

​
Session History

The Sessions tab shows all past sessions for the current project:
Session history with cost tracking

​
Session List

Each session displays:
  • Session ID and slug (human-readable name)
  • Started and ended timestamps
  • Model used (Opus, Sonnet, Haiku)
  • Git branch the session ran on
  • Token usage (input, output, cache creation, cache read)
  • Estimated cost in USD
  • Message count and tool count

​
Session Details

Click a session to view:
  • Full conversation (user prompts and assistant responses)
  • Tool usage patterns (which tools were called and how often)
  • Files changed (complete list with diffs)
  • Cost breakdown (input, output, cache creation, cache read)
  • Timeline of events

​
Cost Tracking

CodeFire calculates cost using model-specific pricing:
ModelInputOutputCache WriteCache Read
Opus$15/M tokens$75/M tokens$18.75/M tokens$1.50/M tokens
Sonnet$3/M tokens$15/M tokens$3.75/M tokens$0.30/M tokens
Haiku$0.25/M tokens$1.25/M tokens$0.30/M tokens$0.03/M tokens
Total cost per session and per project is displayed in the UI.
Use per-project cost tracking to bill clients accurately or monitor your AI spend.

​
How It Works

CodeFire parses Claude Code’s session JSONL files:
  1. Project discovery — Scans ~/.claude/projects/ to find all projects
  2. Session import — Parses .jsonl files to extract metadata and token usage
  3. Live monitoring — Watches active session files for new events
  4. Incremental updates — Reads only new bytes (not the entire file) on each poll

​
Session File Format

Claude Code writes each event as a JSON line: 
{"type": "user", "timestamp": "2026-03-04T10:30:00.000Z", "message": {...}}
{"type": "assistant", "model": "claude-sonnet-4", "usage": {...}, "content": [...]}
CodeFire parses these lines to extract:
  • Session metadata (ID, slug, model, git branch)
  • Token usage (input, output, cache)
  • Tool invocations (name, input parameters)
  • Files changed (from Read, Edit, Write tools)

​
Inactivity Detection

CodeFire considers a session ended if no new events appear for 2 minutes. When a session ends:
  • Live monitoring stops
  • Final token counts and cost are calculated
  • Session is marked as complete in the history
Filter sessions by:
  • Date range — Last 7 days, last month, custom range
  • Model — Opus, Sonnet, Haiku
  • Git branch — Filter by branch name
  • Cost — Sessions above/below a cost threshold

​
Best Practices

Keep an eye on the context meter. If it approaches 200k tokens, consider starting a new session to avoid hitting the limit.
Check which tools the agent called most frequently. Heavy Bash usage might indicate complex workflows; heavy Read usage suggests exploration.
Use CodeFire’s per-project cost tracking to understand spend patterns and bill clients accurately.
Sessions are stored in SQLite (~/Library/Application Support/CodeFire/codefire.db). Query directly for custom reports.

​
Agent Status Bar

When Claude Code is running, CodeFire displays an agent status bar in the terminal panel:
  • 🟢 Active — Claude is running
  • 🟡 Thinking — Claude is processing your request
  • 🔴 Error — Claude encountered an error
This provides quick visual feedback without switching to the Dashboard tab.

Project Management

View sessions grouped by project

Task Tracking

Link sessions to specific tasks

Terminal

Launch Claude sessions from the integrated terminal

Git Integration

See which git branch each session ran on

Build docs developers (and LLMs) love

Get started for freeTalk to us