Run a private portfolio analysis agent. A Cocapn Fleet vessel.
You host this agent 100% on your own Cloudflare account. It makes direct LLM calls with your API key. No third-party servers handle your portfolio data. Open source, MIT license, zero runtime dependencies. Fork-first software.
Most portfolio trackers sell your trading data or lock you into rigid dashboards. This agent was built to be a private, extensible tool that adapts to your strategy, not a platform's. You own every part of it.
- Private Execution: All LLM calls run directly from your Cloudflare Worker with your API key. Your portfolio data never passes through a shared service.
- Personalized Analysis: It builds a persistent knowledge graph based on your specific assets, risk rules, and trade history, enabling more relevant pattern recognition over time.
- Agent Communication: It can use the Fleet protocol to cross-verify signals with other trusted agents you run, without sharing your actual holdings.
- Fork this repository. This gives you the full, unmodified codebase to own and modify.
- Deploy to your Cloudflare account:
gh repo fork Lucineer/activeledger-ai --clone cd activeledger-ai npx wrangler secret put DEEPSEEK_API_KEY npx wrangler deploy - Edit the seed prompts in
src/worker.tsto reflect your portfolio, risk tolerance, and strategies.
Your agent will be live at your own .workers.dev address.
Demo: https://activeledger-ai.casey-digennaro.workers.dev
- Portfolio-focused knowledge graph preloaded with trading and risk management concepts.
- Memory system that connects analysis patterns across your sessions using structural similarity.
- Bring-your-own-key LLM routing: compatible with DeepSeek, SiliconFlow, OpenRouter, and similar providers.
- Fleet protocol support for trusted, privacy-preserving communication between your agents.
- Confidence scoring on analytical outputs.
- Dead-band logic to skip processing when portfolio inputs haven't changed meaningfully.
- Automated pruning of stale, low-confidence nodes from the knowledge graph.
A single Cloudflare Worker. Request state is handled in memory. Long-term pattern data is persisted only to your Cloudflare KV namespace. No external services are required beyond your chosen LLM provider.
The knowledge graph is stored in Cloudflare KV, which has a 1GB limit per namespace and lacks complex querying. Graph traversal is linear, so performance can degrade noticeably beyond a few thousand nodes.
Superinstance and Lucineer (DiGennaro et al.)