Sessions Viewer

English

简体中文
日本語

English

简体中文
日本語

Appearance

Where Antigravity CLI stores session history ​

Antigravity CLI (agy) stores transcript steps under ~/.gemini/antigravity-cli/brain/<conversation-uuid>/.system_generated/logs/. Compare transcript.jsonl with transcript_full.jsonl; neither is guaranteed complete:

~/.gemini/antigravity-cli/brain/<conversation-uuid>/.system_generated/logs/transcript.jsonl

The full layout:

~/.gemini/antigravity-cli/
├── brain/<conversation-uuid>/
│   ├── .system_generated/logs/
│   │   ├── transcript.jsonl        ← rolling window
│   │   └── transcript_full.jsonl   ← nominally complete
│   ├── chunks/
│   ├── scratch/
│   └── media__<timestamp>.<ext>    ← your attachments
└── history.jsonl                   ← the index

Which project a conversation belongs to ​

Nothing in the path says. The index at the root does:

bash
jq -r '[.timestamp, .workspace, .display] | @tsv' \
  ~/.gemini/antigravity-cli/history.jsonl | sort -r | head

Each line is {display, timestamp, workspace, conversationId}. workspace is the project directory and conversationId is the folder name under brain/.

Which transcript file to read ​

Neither is reliably complete. transcript.jsonl is a rolling window compacted at checkpoints, and transcript_full.jsonl, despite its name, gets truncated at checkpoints too. The practical rule is to read whichever is larger.

transcript.jsonl is also not append-only. The CLI rewrites the whole file when it compacts, so anything watching it has to handle the file shrinking as well as growing. A tail that assumes monotonic growth will silently stop updating.

The step record format ​

json
{"step_index":12,"source":"USER_EXPLICIT","type":"USER_INPUT","status":"...",
 "created_at":"...","content":"...","thinking":"...","tool_calls":[{"name":"...","args":{}}]}

source is one of USER_EXPLICIT, MODEL or SYSTEM.

Three things about content will bite you if you treat it as plain text:

  • User input is wrapped in XML. The real message sits inside <USER_REQUEST>…</USER_REQUEST>, followed by a metadata block.
  • Tool results carry a two-line prefix. Every one starts with Created At: …\nCompleted At: …\n before the actual output.
  • Code edits embed a diff. CODE_ACTION steps contain a [diff_block_start] marker followed by a unified diff.

There are no token, usage or model fields anywhere in these records, so cost and token accounting are not available for this agent.

Reading an Antigravity CLI conversation from the terminal ​

Print the steps with their source and type:

bash
jq -r '[.step_index, .source, .type] | @tsv' \
  ~/.gemini/antigravity-cli/brain/<uuid>/.system_generated/logs/transcript.jsonl

Pull out just what you typed, with the XML shell removed:

bash
jq -r 'select(.type=="USER_INPUT") | .content | strings
  | select(contains("<USER_REQUEST>") and contains("</USER_REQUEST>"))
  | split("<USER_REQUEST>")[1] | split("</USER_REQUEST>")[0]' \
  ~/.gemini/antigravity-cli/brain/<uuid>/.system_generated/logs/transcript.jsonl

How do I resume this conversation? ​

Run agy --conversation CONVERSATION_ID from the workspace with the matching CLI installed. The app offers terminal resume, not in-app Antigravity chat.

Source and limitations ​

Implementation: Antigravity CLI adapter.

Reference scope: checked against the Sessions Viewer 0.6.0 adapters on 2026-10-05, not a compatibility guarantee for every CLI release. Shell examples use macOS/Linux paths and require jq (or sqlite3 for opencode). Replace placeholders with real values and use your configured data root. Text extraction does not replay tools, images or the active branch of a session tree. Command regressions use downloadable synthetic samples, not private sessions. Check the capability and evidence matrix for tested scope. Browsing is distinct from resuming or editing data; see privacy and data handling.

This page describes the agy files read by the adapter, not every product named Antigravity. A larger file may preserve more content, but cannot recover content already absent from both files. See missing-session troubleshooting.

Or open it in an app ​

Sessions Viewer picks the larger of the two transcripts, strips the XML wrapper and the timestamp prefixes, renders CODE_ACTION diffs as real diffs, and handles the file being rewritten underneath it. It also reads Claude Code, Codex, Grok Build, Kimi Code, Pi and opencode.