A persistent AI assistant. It retains your conversation and document history across sessions, so you spend less time re-explaining context. Your data is stored in your own Cloudflare account.
Live example: doclog-ai.casey-digennaro.workers.dev
This is a fork-first project. You run your own independent copy.
- Fork this repository.
- Deploy to Cloudflare Workers:
npx wrangler deploy
- Set your API keys as secrets (only you hold these):
npx wrangler secret put DEEPSEEK_API_KEY npx wrangler secret put GITHUB_TOKEN
Your instance will be live in about 10 seconds. You own all data it creates.
Most chat interfaces reset when you close the tab. This is built for work that continues over days or weeks, keeping your project's context accessible.
- Persistent Memory: Conversations and notes are saved to Cloudflare KV.
- Document Tools: Handles summaries, format conversion, and template generation.
- Multi-Model: Defaults to DeepSeek, but works with any OpenAI-compatible API.
- Your Data: You can export all your data as plain JSON at any time.
- Built-in Safeguards: Includes per-IP rate limiting.
- Fleet-Compatible: Supports the Cocapn Fleet protocol for coordination with other agents.
A single Cloudflare Worker file with zero runtime dependencies. All state is stored in your Cloudflare KV namespace.
Cloudflare KV has a 25MB limit per key. In practice, a single conversation thread can hold roughly 50,000 words before you may need to archive or prune older messages.
MIT License.
Attribution: Superinstance and Lucineer (DiGennaro et al.).