TinyPelican

AI social copilot. A local-first relationship reasoning engine: parse chat logs, build living contact profiles, detect relationship signals, and generate follow-up todos, schedules, and reply drafts.

🌐 English · 中文

---

⚠️ Read this first

This project is not a turnkey personal-WeChat assistant:

• ❌ It does not silently read WeChat in the background. Chat logs are ingested by automatic clipboard detection: you copy a chat, TinyPelican detects and archives it automatically — or via an experimental iLink bot channel.
• ❌ It does not claim official WeChat integration or real-time message sync for all scenarios.
• ✅ It does provide the open relationship-reasoning core: chat parsing, contact profile extraction, intent/todo/schedule generation, proactive reminders, and reply drafting.

If you are interested in the reasoning engine and the product idea, you are in the right place.

What TinyPelican does

Traditional personal CRMs like Monica require manual data entry. TinyPelican turns copied chat logs into a self-growing relationship memory:

Capability	Description	Status
Chat ingestion	Automatic clipboard detection of WeChat chat logs + experimental iLink channel	✅ Live
Contact profiles	One profile per contact, continuously extracted from conversations	✅ Live
Relationship reasoning	Importance, recent context, preferences, boundaries, social goals	✅ Basic
Intent extraction	Tasks, deadlines, schedules, waiting-for-reply	✅ Basic
Proactive reminders	Deadline reminders, do-not-disturb, neglected-relationship nudges	✅ Basic
Reply drafts	Relationship-aware suggestions, one-click fill, never auto-sent	✅ Live
Agent execution	DSH Agent with structured Skills and auditable event stream	✅ Live
Spoken diary via bot	Tell the WeChat bot; memory is settled by the Agent	🚧 In progress
More social channels	QQ, Feishu / DingTalk, Telegram, email, calendar	⏳ In development
Wearable activity import	AI glasses / cameras / watches as future memory inputs	⏳ Planned

Why it's different

	TinyPelican	Monica	General AI assistant
Data entry	Automatic from copied chat logs	Manual	None
Relationship profiles	AI-extracted, continuously updated	Manual forms	No persistent memory
Proactive reminders	From chat commitments and relationship signals	User-set reminders only	No
Reply drafts	Relationship-aware	No	Generic
Local-first	Yes, BYO model key	Self-host option	Cloud

More channels are coming — contributors wanted

Current input is WeChat-first (clipboard auto-detection + experimental iLink). More social channels are under active planning/development:

• QQ, Feishu / DingTalk, Telegram, email, calendar
• Wearable activity import (AI glasses / cameras / watches)

This project is maintained by an undergraduate student, so multi-channel support will take time. Every channel follows a simple connect / stop / send contract — if you are interested in channel adapters, Agent Skills, or the relationship-reasoning core, you are very welcome to join as a contributor. See CONTRIBUTING.md.

Quick start

Requirements: Windows, Node.js 18+, DSH WebUI (http://127.0.0.1:3080), one OpenAI-compatible model provider.

git clone https://github.com/Oii6111/TinyPelican.git
cd TinyPelican
npm install
Copy-Item config.example.json config.json
npm run daemon

Open http://127.0.0.1:18791 for the WebUI.

Desktop shell:

cd app
npm install
npm start

Commands:

npm run daemon      # core service, clipboard, WeChat, scheduler, Agent queue
npm run server      # WebUI/API only
npm test            # tests
npm run check       # syntax checks
npm run remind:dry  # preview reminder rules

Architecture

Chat logs (clipboard / experimental iLink channel)
        ↓
Parse, deduplicate, resolve contact
        ↓
SQLite contact & chat memory + structured tasks/schedules
        ↓
Structured APIs
        ↓
DSH Agent + Skills
        ↓
Drafts / pending intents / tasks / reminders / schedules
        ↓
Web dashboard / Electron overlay / WeChat & Bark notifications

Repository map

app/                 Electron desktop shell
agent/               Agent.md, Skills, DSH event stream
core/                APIs, Agent, SQLite, ingestion, reminders, channels
dashboard/           Native ES Module WebUI
tests/               Unit and integration tests

Runtime data (SQLite, tasks, schedules, intents, credentials, logs, chat records) is gitignored.

Roadmap

1. Make contact memory, proactive reminders, and reply drafts a daily habit.
2. Improve intent accuracy with human-in-the-loop feedback.
3. Group-chat model and wearable activity import adapters.
4. Add more channel adapters (QQ, Feishu / DingTalk, Telegram, email, calendar).

Contributing

Contributions are welcome — see CONTRIBUTING.md. We keep a good first issue label for newcomers.

License

MIT License — free for personal learning and open-source use. See LICENSE.

---

⭐ If this project inspires you, a star helps the project grow. Thank you!

Repository: github.com/Oii6111/TinyPelican