Arrive with context
Open a person and see recent messages, accepted memories, relationship notes, and open loops together.
v0.2 developer preview
Your conversations know you. Now your laptop does too.
Weft turns WhatsApp and Telegram history into useful relationship memory, ready-to-review replies, and the follow-ups you meant to make. Your messages stay on your machine.
- WhatsApp + Telegram
- Local models
- You approve every send
Actual Weft interface · Demo Provider
01 / Product
One calm place for the context behind every relationship.
Weft does the remembering around your conversations, then puts every suggestion in front of you before it becomes a memory, task, or sent reply.
A real product walkthrough
See the relationship memory surface in 64 seconds.
Recorded from the v0.2 Demo Provider.
No personal messaging data is shown.
02 / Memory loop
Useful before you type.
Draft on your hardware
A local OpenAI-compatible model prepares grounded reply suggestions from the context you control.
Keep the human decision
Review generated memories, open loops, and drafts. Nothing becomes durable or gets sent without you.
03 / Trust model
Local is the architecture, not a setting.
Weft is built for conversations that should not become someone else's training data. Storage, inference, connectors, and approvals run around a local trust boundary.
- Storage
- Messages, memories, drafts, and audit history live in local SQLite.
- Inference
- Chat and embeddings use OpenAI-compatible endpoints you configure.
- Agent access
- MCP clients only see conversations you explicitly mark visible.
- Sending
- Drafts stay drafts until you review and approve them.
04 / Source code
Follow Weft while the product takes shape.
Weft is early pre-release software and is not ready for a dependable local install yet. The repository is public for anyone who wants to inspect the architecture, track progress, or contribute.
GitHub repository
michaellee8/weft
Browse the React management UI, Rust daemon, local connector code, architecture notes, and current development history.
Check out the source code