Summary
Codeburn currently supports Claude Code, Codex CLI, and Cursor (per --provider flag in 0.9.2). Adding Gemini CLI would close the loop for the most common multi-provider setup.
Why
Many users run all three concurrently (Claude Code as primary, Codex for code review, Gemini for research / large-context work). Codeburn already gives the cleanest TUI in this space, so the lack of Gemini means we either compromise on UX or run a second tracker just for Gemini cost visibility.
Where Gemini CLI stores logs
Path:
~/.gemini/tmp/<project>/chats/session-<timestamp>-<hash>.jsonl
In my install I have 83 active session files across multiple projects.
Tested with:
- Gemini CLI:
0.39.1
- macOS (Apple Silicon)
Format (verified, real sample, sanitized)
First line is session metadata, then JSONL with messages:
{"sessionId":"<uuid>","projectHash":"<sha256>","startTime":"2026-04-27T01:55:42.605Z","lastUpdated":"2026-04-27T01:55:42.605Z","kind":"main"}
{"id":"<uuid>","timestamp":"2026-04-27T01:55:45.373Z","type":"user","content":[{"text":"..."}]}
{"$set":{"lastUpdated":"2026-04-27T01:55:45.373Z"}}
{"id":"<uuid>","timestamp":"2026-04-27T01:55:50.123Z","type":"model","content":[{"text":"..."}]}Message type values seen: user, model, plus stream-update lines with $set.
Pricing references
For cost calculation, Gemini API pricing is published at https://ai.google.dev/pricing. Models I use most often:
gemini-3.1-pro-previewgemini-2.5-progemini-2.5-flash
Notes
I tested splitrail (which advertises Gemini support) and found two related issues there:
- It looks for
*.json, but Gemini CLI writes *.jsonl (mismatch causes 0 sessions detected).
- After renaming a sample to
.json, splitrail registers the analyzer but parses 0 messages, suggesting the schema check expects an older format.
Sharing in case it's useful when designing the codeburn parser.
Happy to test pre-release builds against my 83 sessions if helpful.
Summary
Codeburn currently supports Claude Code, Codex CLI, and Cursor (per
--providerflag in 0.9.2). Adding Gemini CLI would close the loop for the most common multi-provider setup.Why
Many users run all three concurrently (Claude Code as primary, Codex for code review, Gemini for research / large-context work). Codeburn already gives the cleanest TUI in this space, so the lack of Gemini means we either compromise on UX or run a second tracker just for Gemini cost visibility.
Where Gemini CLI stores logs
Path:
In my install I have 83 active session files across multiple projects.
Tested with:
0.39.1Format (verified, real sample, sanitized)
First line is session metadata, then JSONL with messages:
{"sessionId":"<uuid>","projectHash":"<sha256>","startTime":"2026-04-27T01:55:42.605Z","lastUpdated":"2026-04-27T01:55:42.605Z","kind":"main"} {"id":"<uuid>","timestamp":"2026-04-27T01:55:45.373Z","type":"user","content":[{"text":"..."}]} {"$set":{"lastUpdated":"2026-04-27T01:55:45.373Z"}} {"id":"<uuid>","timestamp":"2026-04-27T01:55:50.123Z","type":"model","content":[{"text":"..."}]}Message
typevalues seen:user,model, plus stream-update lines with$set.Pricing references
For cost calculation, Gemini API pricing is published at https://ai.google.dev/pricing. Models I use most often:
gemini-3.1-pro-previewgemini-2.5-progemini-2.5-flashNotes
I tested splitrail (which advertises Gemini support) and found two related issues there:
*.json, but Gemini CLI writes*.jsonl(mismatch causes 0 sessions detected)..json, splitrail registers the analyzer but parses 0 messages, suggesting the schema check expects an older format.Sharing in case it's useful when designing the codeburn parser.
Happy to test pre-release builds against my 83 sessions if helpful.