Sessions

Sessions are the core of Metateam. Every AI coding session can be captured, stored, and searched.

For trust boundaries and what gets stored/uploaded, read Security and Privacy.

Search Sessions

$ metateam session search [QUERY] [OPTIONS]

Examples:

$ metateam session search "session hooks" -n 3
$ metateam session search --when 7d -n 5
$ metateam session search --when today --tags persona-rivet

--when supports: today, yesterday, 7d, 2w, 1m, 1y, exact dates, and ranges.

Load a Session

$ metateam session show <SESSION_ID>

Session ID accepts git-style prefixes (8+ hex chars).

Useful options:

$ metateam session show <SESSION_ID> --full
$ metateam session show <SESSION_ID> -s 2000

Load the Most Recent Session

$ metateam session last

Optional output controls:

$ metateam session last --full
$ metateam session last -s 2000

Session Metadata

$ metateam session info <SESSION_ID>

Use --title for title-only output and -j for JSON.

Delete a Session

$ metateam session delete <SESSION_ID> [-y]

Pass -y to skip confirmation.

Upload Sessions

$ metateam session upload [SOURCE_OR_TARGET] [FILE]

Examples:

$ metateam session upload claude
$ metateam session upload grok
$ metateam session upload all
$ metateam session upload 82bbfb7c
$ metateam session upload claude file.jsonl

SOURCE_OR_TARGET can be a provider (claude, codex, agy, pi, opencode, grok, all), a session prefix (8+ hex), or a file path. Grok discovery reads only each session's authoritative updates.jsonl and uses its sibling summary for cwd safety checks.

Stats

$ metateam session stats

Antigravity (agy) transcripts are under ~/.gemini/antigravity-cli/brain/<conversation-id>/.system_generated/logs/transcript_full.jsonl. METATEAM_AGY_ROOT overrides ~/.gemini/antigravity-cli. Metateam reads the full transcript because transcript.jsonl truncates large fields. The workspace's current conversation comes from cache/last_conversations.json. A clear creates a new conversation directory, which Metateam checks without the sibling-file cache. Catch-up upload uses conversation_summaries.db to find workspaces for older conversations. A missing full transcript is not replaced with the truncated sidecar.

The full agy transcript has token counts but no model name per step. Metateam displays the model observed by lifecycle hooks. It leaves cost unavailable because it cannot attribute historical tokens to models reliably.