Odysseus AI

Odysseus AI is an AI workspace designed with privacy at its core, positioned as a "private environment for deep research and autonomous agent workflows," primarily addressing the issue of data control when using AI for material organization and research in confidential or highly sensitive scenarios.

Main Uses

• Transform users' own source materials (PDF, Markdown, TXT, URLs, pasted notes, sources already saved in the workspace) into structured research reports with traceable sources.
• Run autonomous agent tasks and memory workflows in an isolated environment.
• Provide automated intelligence organization capabilities for highly sensitive analysis scenarios, with the official emphasis on "no data leakage."

Main Capabilities

• Source-traceable reports: Generated reports come with a strict citation-style source table, making it easy to trace each item back to the original material.
• Autonomous agents: Research agents can be run on complex questions, automatically organizing URLs, PDFs, and notes into structured, traceable briefs; real-time intervention is supported, allowing interruption, redirection, and refinement (Live agent steering).
• Model routing and BYOK: Managed routing is provided through OpenRouter, while also supporting users bringing their own OpenRouter API Key (BYOK) so they can control billing and model selection themselves; the official statement says zero data retention (ZDR) routing is supported.
• Custom MCP tools: Custom MCP server context can be connected to extend the tools and information available to agents.
• Isolated memory: Saved sources, notes, and past reports can serve as context for subsequent agent runs, gradually building a private knowledge base.
• Controlled file access: Agents have no access by default and can only access URLs, files, and tools explicitly authorized by the user in the current session; they cannot read the broader workspace.
• No telemetry by default: The official description states that the product experience does not rely on behavioral tracking or invasive analytics.

How to Use

1. Securely upload PDF, Markdown, and TXT files and notes in the isolated workspace.
2. The engine organizes the selected sources into a highly structured context window.
3. Set research parameters for the autonomous agent and define strict execution limits.
4. Review the generated brief, verify the source table, and audit the agent's actions.
5. Keep the generated insights in isolated memory for subsequent tasks.

Applicable Scenarios

• Academic and technical literature reviews: Import technical PDFs, standards documents, and notes to produce conclusions with strict citation source tables.Structured overview.
• Competitive and product benchmarking: Add competitor documents, pricing pages, and internal notes to generate a benchmarking brief with traceable evidence.
• Market intelligence and OSINT: Store analyst reports, press releases, and market data, run autonomous loops, and produce reviewable and verifiable market briefs.
• Due diligence and investment research: Securely upload due diligence materials and public documents to generate due diligence briefs that can be traced item by item back to the underlying materials.

Limitations and Information Not Provided

• The official statement says it is built on an isolated cloud architecture, and that this infrastructure "lays the foundation" for self-hosted and on-premises deployment paths, but it is currently not a self-hosted product; local model support is still evolving, and at this stage model privacy is safeguarded through OpenRouter BYOK.
• Agents cannot automatically access files; they can only use the documents, notes, and context explicitly provided by the user for that session.
• Hosted model requests are processed through OpenRouter and the selected downstream model providers; the official statement says user data is not used to train models, and BYOK can be chosen to route through one's own key.
• The official statement mentions that there is a public GitHub repository on the homepage, but the specific license, branch status, and which components are public must be based on the repository, and details are not given in the materials.
• Subscription plans include different tiers of monthly research report counts, estimated hosted agent run counts, number of saved sources and notes, workspace storage, and so on; standard reports and hosted agent runs share the included hosted AI budget, the actual number of runs varies with task complexity, and BYOK can be used to exceed the included hosted usage. Annual payment is the default, monthly payment can be switched to at any time, and a 30-day money-back guarantee is claimed.
• The materials do not provide facts such as the specific price figures for each subscription tier, the specific list of available models, or details of downstream providers other than OpenRouter.