Odysseus AI

Inspiration

What it does

How I built it

Challenges I ran into

Inspiration

Most AI tools route sensitive documents, competitive research, and proprietary notes through third-party training pipelines with little transparency. We wanted a private AI workspace built specifically for deep research and autonomous agent work, so researchers, analysts, and builders would not have to trade privacy for capability.

What it does

Odysseus AI is an isolated, source-traceable workspace for deep research and agent execution. Upload PDFs, Markdown, and text files, then deploy autonomous research agents that turn those sources into structured, citation-backed reports. A persistent memory layer keeps saved sources, notes, and prior reports available as reusable context, so new research tasks build on earlier work instead of starting from zero. Model access runs through managed OpenRouter routing, with an optional bring-your-own-key (BYOK) mode for teams that want direct control over which models process their data. Common workflows include academic and technical literature review, competitor and product benchmarking, market intelligence and OSINT, and due diligence for investment research.

How we built it

Odysseus runs as an isolated managed-cloud workspace built on Cloudflare's infrastructure, with Stripe handling billing and OpenRouter routing model calls across providers including DeepSeek. Agent workflows connect through Model Context Protocol (MCP) tools, letting the system pull in sources, URLs, and files under strict, user-defined execution limits. Every workspace enforces zero-data-retention (ZDR) routing, so files, prompts, and outputs are never used to train models, and agents only access what a user explicitly authorizes for that session.

Challenges we ran into

The hardest problem was balancing autonomy with control: agents need enough freedom to run multi-step research loops, but users need a guarantee that those agents will only touch the sources, URLs, and tools authorized for that specific session. Designing a memory system that stays useful across sessions, without ever compromising workspace isolation, took several iterations to get right.

Accomplishments that we're proud of

We shipped an end-to-end workflow that takes a user from raw sources to a finished, cited research report in five steps: connect, structure, delegate, review, and continue. The workspace keeps zero telemetry by default and full source traceability in every generated report, which was a hard requirement from day one.

What we learned

Privacy and usability do not have to be at odds. By committing early to zero-data-retention routing and strict, explicit agent permissions, we were still able to deliver fast, structured research and benchmarking output comparable to more open, less private AI tools.

What's next for Odysseus AI

We are expanding the MCP tool ecosystem, adding more structured use cases such as due diligence and competitor benchmarking, and continuing to strengthen the isolation guarantees that make Odysseus a private alternative to mainstream AI research tools.

Accomplishments that I'm proud of

What I learned

What's next for Odysseus AI

Built With

• cloudflare
• deepseek
• mcp
• openrouter
• stripe