Turn company knowledge into grounded deliverables.
Donna builds a structured, searchable knowledge layer from existing documents, then helps agents regenerate Markdown deliverables from memory, context and templates.
Local-first by design. Source-grounded by default. Flexible across OpenAI, Anthropic, OpenAI-compatible endpoints and local models.
Regulated environments Donna is designed to run without third-party desktop clients such as Claude Desktop or similar applications, which are often not approved in regulated organizations. It can operate locally or on controlled infrastructure, helping teams align with ANSSI security expectations and DINUM sovereign digital tooling principles. RGAA-minded DSFR-inspired On-premise ready Sovereign AI compatible MCP local tools
The model is no longer the center.
Donna follows the Karpathy pattern: useful AI systems are shaped by context, memory, tools and skills around the model. The model remains pluggable; the durable advantage is the knowledge layer and the procedures around it.
That is why Donna is presented as one product: a local application that gives agents grounded memory, traceable context and callable capabilities.
A product layer for living knowledge.
Wiki
A structured, searchable knowledge layer built from existing documents. Pages remain traceable back to their source, and the wiki can be exposed through MCP so other agents on the same machine can read from it.
Build
A build agent combines exactly three inputs — memory, context and a Markdown template — to regenerate a deliverable. When sources change, the user can regenerate instead of rewriting the same document by hand.
Pilot
A chat interface connects external MCP tools and reusable skills, then composes multi-step work without asking the user to script each step manually.
One local core. Multiple surfaces.
Sources enter from the left. Donna turns them into memory, context and buildable deliverables in the center. People and agents consume the result on the right. The model provider is swappable underneath.
See how knowledge connects.
The graph view makes the knowledge layer visible. It helps users inspect relationships between pages and sources, spot isolated content and understand where a deliverable gets its grounding.
For teams that need auditability, this is the difference between a black-box assistant and a navigable source-backed system.
Let the agent compose the work.
The chat is not only a conversation surface. It is where MCP tools are connected, skills become reusable named procedures, and multi-step work is triggered without hand-written scripts.
The user stays in control while Donna reads, builds, checks and routes work through the available tools.
Bring the model that fits the job.
Donna is explicit about multi-provider execution: cloud models, OpenAI-compatible endpoints and local models can all fit the same product pattern.
Runs where your knowledge already lives.
Donna is not positioned as a SaaS. The product is designed to run on the user’s machine or infrastructure, with local indexes, local workspaces and optional local model execution.
For sensitive environments, the local-first model is not a footnote. It is the feature.
Install Donna. Build from source-backed memory.
A practical path to AI deliverables that can say where they came from — and when the source is missing.