The leading nonprofit for AI in public services
We build the open-source technology that lets public services around the world deploy AI agents to better support their citizens.
Everyone who needs help can receive it.
The operating system of the agentic government
From employment to health to administrative procedures, the Bayes Platform lets public services deploy AI agents in the service of their citizens.
1Safe
It follows your protocols rather than the model's defaults.
2Effective
It is measured in production, on field results.
3Sovereign
It runs on your infrastructure, with your data, in open source.
AI to empower people in need and those who help them
We believe AI can bring everyone personalised help, at scale. That's why we build it open-source and as a nonprofit, so the most advanced technology goes to the most pressing needs.
Introducing the Bayes Platform
The first open-source AI agent platform built for public services. Here is what it does, in the real product.
The agent is tied to no model and no supplier. Changing either means retesting and recalibrating the harness, with no business logic to rewrite, and open models remain an option when the case calls for one.
Managed cloud or your own infrastructure, from a single Docker image. The retention period is yours, and your data never trains a third-party model. GDPR, AI Act, SOC 2 Type II, ISO 27001 and HDS compliance.
The code is MIT-licensed: you deploy it, change it, and keep it if Bayes disappears. What one institution funds benefits the next, with no licence to buy.
The agent answers from your handbooks, protocols and official references, never from the open web. Every answer cites the document and the extract it used, and the platform enforces your rules when the model skips them.
A model alone does not make a public-service agent. What matters is everything around it: sources, rules, evaluation, monitoring. The Bayes Platform is that harness, built for public service and nothing else.
Your frontline teams create and tune their own agents, with no code. Every change writes a draft, and nothing reaches the public that a human has not published.
Evaluation campaigns replay conversation and extraction datasets against a chosen version. Reviewers rate the answers blind, an LLM judge gives the scale, and two runs compare side by side.
Every exchange is filed under the life situations you defined, and the topics your answers cover badly surface on their own. You complete the corpus, re-evaluate and publish: the agent improves with every version.
Each agent is exposed over API and MCP, an open standard: the assistant or portal your institution already runs calls it as a tool. A widget or a public page when that is simpler. Nobody changes screen.
In the press
A few selected press features about our mission and work.