multi-session

Multi-Session Context Example

Demonstrates how lance-context uses MemWAL (Memory Write-Ahead Log) to handle concurrent, multi-bot context storage with automatic shard isolation.

What this shows

• Multiple bots (coding-assistant, research-assistant) writing to the same dataset
• Automatic sharding by (bot_id, session_id) — each pair gets its own WAL region
• No contention between concurrent writers targeting different shards
• Time-travel with checkout() across all shards

How MemWAL sharding works

When you call ctx.add(..., bot_id="X", session_id="Y"), lance-context:

1. Hashes (bot_id, session_id) into a deterministic UUID v5 shard ID
2. Routes the write to that shard's WAL region
3. Periodically compacts WAL regions into optimized Lance fragments

This means multiple agents can write simultaneously without coordination — each writes to its own isolated WAL region.

Run

pip install lance-context>=0.2.5
cd examples/multi-session
pip install -e .
multi-session-demo

Expected output

Created context store at .artifacts/multi_session_XXXXXXXX.lance
After coding session: version=N, entries=5
After research session: version=N, entries=9
After second coding session: version=N, entries=14
Rolled back to version N: entries=5
Restored to version N: entries=14
All writes were sharded by (bot_id, session_id) via MemWAL. Each pair wrote to its own WAL region independently.