Provider

Claude Code

Claude Code writes JSONL transcripts to disk. Budi tails these files in real-time for session, token, and cost data. No configuration needed — just install budi and it finds the transcripts automatically.

Live since v8.0.0

Data sources

Local JSONL transcripts (primary)

Claude Code writes structured JSONL transcript files that include every message in a conversation — user prompts (content not read by budi), assistant responses, tool calls, and token usage metadata. Budi tails these files as they're written.

Path: ~/.claude/projects/*/

What's parsed

Input and output token counts per message
Cache creation and cache read token counts
Model name and version (e.g. claude-sonnet-4-6, claude-opus-4-6)
Dollar cost (computed from token counts via budi's pricing manifest)
Session boundaries and conversation timestamps
Repo, branch, and working directory from project path

Caveats

Zero-config provider

Claude Code is the simplest provider to set up — budi auto-discovers the transcript directory and starts tailing immediately. No API keys, no PATs, no configuration.

Cost is computed from token counts using budi's pricing manifest, not from a billing API. This means cost tracks Anthropic's published pricing — if your org has custom pricing, the numbers may differ.
Budi reads only the structured metadata (tokens, model, timestamps) — it does not read your prompts or the AI's responses.

How to verify

After installing budi, start a Claude Code session and send a few messages. Then run:

budi doctor

Look for the Claude Code tailer in the output:

✓ Daemon running (pid 12345, port 7878)
✓ Claude Code tailer: active
  Path: ~/.claude/projects/
  Last event: 2s ago
✓ All checks passed.

To confirm session data is flowing:

budi stats projects