DEVELOPERS
“Three agents, kept firmly in their lane.”
Useful, perfected AI where it actually helps, not AI for its own sake. Strawbay’s agents work inside the console with a deliberately constrained reach: the right knowledge, the right tools, and strict isolation per customer. People stay in control, and every change to production is still a human, authorised step.
“We did not want AI for its own sake. We wanted useful AI that actually helps people in the console, and the interesting work is the guardrails around it, not the model.”
Rickard Elmqvist, CEO, Strawbay · Read the article
Three agents
Three agents, three jobs
Anna
Anna assists. She answers questions and helps users get things done, working from the knowledge base and read access to what each user is allowed to see.
Saga
Saga teaches and enables. She guides users through the platform and how integrations work, so teams get productive faster.
Barry
Barry builds. He can create and change integrations, but every write runs through plan mode and explicit approval before anything happens.
Useful and perfected, and governed
How we keep the AI in its lane
A governed knowledge base
Agents answer from a continuously maintained knowledge base: product docs, the DSL catalog, API specs and curated uploads. They do not improvise from the open internet.
No customer data in the store
Knowledge is sanitized with defense in depth, secrets and personal data are scrubbed before anything is stored. Customer data never enters the shared store; agents read your live data through scoped tools only when they need it.
Tools through a governed layer
Agents reach the platform through a governed tool layer, our approach to the Model Context Protocol. The model only ever sees the capabilities its role is permitted to use, never open-ended access.
Approval and plan mode
A permission gate sits in front of every action. Safe ones run; sensitive or destructive ones stop for explicit approval. Barry proposes a documented plan, you approve, then it executes step by step.
Isolated per customer
Every conversation is bound to the user and their tenant. Knowledge, data and tools are scoped per connector, and per-customer budgets are enforced fail-closed.
Inference stays in the EU
Models run on Vertex AI in europe-west1, the same region as the rest of the platform, so prompts, completions and tool data stay in the EU on the default path.
AI speeds people up; it does not replace their judgement. For how we build and release, see our Development Methodology.