Private AI for WhatsApp and Telegram

Your social second brain, running on your laptop.

Weft helps you draft replies, remember people, and follow up with friends, customers, and partners without uploading conversations to cloud AI.

Run it locally See what it does

No cloud upload Local LLMs Human-approved sends

Weft interface preview showing local AI suggestions, memories, and draft replies

Local AI, not cloud CRM leakage

Draft replies before you need them.

Local models are slower than cloud APIs, so Weft prepares useful suggestions in the background. When you open a conversation, draft replies, relationship briefs, memory candidates, and open-loop candidates are already waiting for review.

The chat completion queue defaults to one request at a time, so your laptop stays responsive while local inference does the work.

Reply with context

Weft drafts replies from recent messages, accepted memories, and retrieval over your local relationship history.

Remember what matters

Preferences, birthdays, business needs, sensitive facts, and useful summaries become reviewable memory candidates.

Never drop a follow-up

Open loops capture promises, customer requests, and pending replies before they disappear into the scrollback.

Read the room fast

Short relationship briefs remind you who someone is, what happened recently, and how to respond with care.

Built for real message overload

For friends, customers, and the details you keep forgetting.

For people with busy social lives

Remember preferences, plans, birthdays, sensitive topics, and promises before you reply.

For small businesses

Keep customer context close without sending proprietary chats, pricing, or relationship history to a cloud CRM assistant.

For careful operators

Review every generated memory, open loop, and draft. Weft helps, but it does not send messages behind your back.

Privacy model

Your conversation data stays on your computer.

Weft stores data locally, talks to local inference endpoints you configure, and gates agent access behind visibility controls.

Local SQLite database for messages, memories, drafts, and audit history.
Local OpenAI-compatible chat endpoint for reply and memory generation.
Local OpenAI-compatible embedding endpoint for retrieval.
Loopback daemon defaults and first-run setup sessions.
Optional MCP access for local agents, kept secondary and permissioned.

Run the pre-release

Try Weft locally.

Use the Demo Provider first, then connect Telegram or WhatsApp when you are ready to bring in real messages.

npm ci cargo run -p weftd -- --dev --seed-mock-connector npm run dev