Quick start: a synthetic local Lake

This quick start was executed against Transcript Lake 0.2.0 on 2026-08-24 with DuckDB 1.5.5. It used a synthetic Claude-format fixture under an isolated HOME; no real transcript was opened and no credential was used.

1. Build and select an isolated root#

cargo build
export HOME=/private/tmp/lake-demo/home
export LAKE_DATA=/private/tmp/lake-demo/lake

Create a synthetic ~/.claude/projects/-tmp-demo-app/<session>.jsonl fixture containing user, assistant, thinking, tool-call, and tool-result records. Use invented values only. The executed fixture used session id 0a1b2c3d-1111-4222-8333-444455556666 and project /tmp/demo-app.

2. Start the foreground stream#

transcript-lake stream

The retained stream log showed catch-up followed by a real filesystem append:

text

2026-08-24T23:11:40.372Z stream start roots=1
2026-08-24T23:11:40.432Z stream catch-up files=1 streamed=1 failures=0 ms=38
2026-08-24T23:11:47.780Z stream event paths=1 first=/private/tmp/lake-demo/home/.claude/projects/-tmp-demo-app/0a1b2c3d-1111-4222-8333-444455556666.jsonl
2026-08-24T23:11:47.853Z stream commit files=1 failures=0 ms=50

Run the stream under a supervisor for continuous use. Stop it with SIGINT or SIGTERM; the process writes final stream state and releases its writer lease.

3. Inspect the result#

In another terminal with the same HOME and LAKE_DATA:

transcript-lake sources
transcript-lake status

The executed source discovery was:

text

claude: transcripts, 1 files
  /private/tmp/lake-demo/home/.claude/projects
codex: not found, 0 files
omp: not found, 0 files
droid: not found, 0 files
kimi: not found, 0 files
hooks: not found, 0 files

The status snapshot reported one 3,352-byte Claude partition, one healthy cursor, a running stream with zero failures, and Lake data fresher than the absent Oko index.

4. Verify the privacy boundary#

The stored synthetic user event contained these markers:

json

{"runtime":"claude","event_type":"user","text":"Deploy fails. I set [masked:assignment:37:5be474f8] and the provider token is [masked:token:36:e3bebba1] - can you check the config?"}
{"runtime":"claude","event_type":"user","text":"Also scrub this pasted blob before we archive: [masked:entropy:44:d0fa7c09]"}

A literal search for all three invented plaintext values across the isolated lake/ tree returned zero matches in the cursor, partition, stream status, and Oko export files. This demonstrates the three documented masker classes for that fixture; it is not a claim that arbitrary text is anonymized.

5. Read and derive#

With DuckDB 1.5.x on PATH:

transcript-lake sessions --limit 20
transcript-lake show 0a1b2c3d-1111-4222-8333-444455556666 --include all
transcript-lake search "Rotate the leaked key"
transcript-lake compact --source claude

The executed compaction reported:

text

claude: ndjson 3352 bytes -> parquet 5122 bytes (/private/tmp/lake-demo/lake/parquet/runtime=claude/events.parquet)

NDJSON remains authoritative; Parquet is rebuildable. Continue with the live-append walkthrough, masking audit walkthrough, and CLI reference.