Business truth is scattered across databases, files and SaaS tools, so every person and every AI assistant ends up with a different answer. AgentBrain connects those sources, keeps knowledge with its provenance, enforces who may see what, and serves the same permission-aware context to people and AI over MCP and API.
Policies sit in shared drives, numbers sit in databases, and decisions sit in chat threads. When AI assistants read from all of it without rules, they mix outdated documents with current ones and reach data their user should not see. AgentBrain gives that knowledge one governed home.
From data connectors to permission-aware MCP access, AgentBrain is the data and context layer that people and AI assistants share.
Connect the systems where business data already lives and query or extract from them under the same permission model.
Learn moreStore documents and media in folders with permissions, then find them with full-text search refined by semantic ranking.
Learn moreModel entities and relationships across your knowledge, with the evidence behind each fact kept next to it.
Learn moreOne permission model covers folders, knowledge, media, connectors, graph entities and workflows, and important actions land in an append-only audit log.
Learn moreMove and reshape business data with scheduled or webhook-triggered workflows: extract, transform, load, and feed results back into the knowledge base or graph.
Learn moreAI assistants reach governed context through an MCP server that asks the backend to check permissions before returning anything.
Learn moreThis walk-through shows the answer flow AgentBrain is designed for, on a synthetic dataset. It is a design illustration, not a recording of a live system.
“What is our current refund window for annual plans?”
Permission check
The assistant asks over MCP. The backend checks which folders and documents this support agent's role may read.
Authorized sources
Two documents qualify: Refund Policy v3 and Refund Policy v4. A finance-only pricing memo is excluded.
Effective date
v4 is in effect from 2026-07-01; v3 ended on 2026-06-30. The answer uses v4 and says so.
Answer with citation
“30 days from purchase.” Cited: Refund Policy v4, section 2, effective 2026-07-01.
Conflict handling
If two current documents disagree, the answer shows both sources and does not pick one silently.
What the assistant may not do
It cannot read the finance memo, change the policy, or issue a refund. It can only retrieve what the user may already see.
What AgentBrain enforces today, what depends on your deployment and providers, and what is still planned. See the Trust page for status labels.
They are separate products with separate licenses. You can buy either one alone, and connect them when your agents need governed business context.
Connect sources, search and manage knowledge, model it in a graph, govern access, and give AI assistants permission-aware context over MCP or API.
Licensed per AgentBrain instance. See AgentBrain pricing.
AgentBrain pricingRun and coordinate AI agents for teams. Dewee is a separate product with its own license, sold on dewee.sh.
Pricing and terms are published only on dewee.sh.
Dewee pricingDewee agents can read AgentBrain knowledge through the AgentBrain MCP server, which applies AgentBrain permissions to every request.
Requires configuration per deployment. A packaged, tested integration and any bundle price are not available yet.
Discuss an integrationAgentBrain is the governed context layer. Agent runtimes and chat channels are integrations, and each one below is labelled with what exists today.
Connect any MCP-capable assistant to the AgentBrain MCP server for permission-checked search and retrieval.
Separate agent execution product. Connects through MCP with per-deployment configuration.
Agent runtime from the same team. AgentBrain does not depend on it; it can connect as an MCP client.
Workflows can include agent task and LLM call steps. Agent tasks over MCP are not supported yet.
Workflows can send outbound notifications. AgentBrain does not run inbound chat bots.
No native channel integration in AgentBrain today.
AgentBrain comes from the same team behind practical agent tooling, an open agent runtime and AI-assisted design workflows.
From scattered sources to governed context
Connect databases, sheets, storage and documents to AgentBrain.
Set organization, department and resource permissions, and choose AI providers and egress modes.
Give people search and AI assistants permission-aware context over MCP and API.
AgentBrain is the governed business data and knowledge plane. It connects data sources, stores and searches knowledge, keeps provenance on graph facts, enforces organization and resource permissions, and serves permission-aware context to people and AI assistants over MCP and API.
No. Dewee runs and coordinates agents. AgentBrain governs the data and context those agents, and people, are allowed to use. They are separate products you can buy alone or connect.
Grants go to users, roles and groups, scoped by organization and department, on folders, knowledge, connectors, graph entities and workflows. Requests from MCP clients are checked by the backend, and important actions are written to an append-only audit log.
AgentBrain runs as a managed cloud service today. Private deployments on your infrastructure are scoped per project. Air-gapped operation is not supported.
Content goes to whichever AI model or embedding provider you configure. You can block external AI traffic per organization or use self-hosted OpenAI-compatible models. See the Trust page for the full data flow.
Book a demo. We will look at your sources and permission needs, then scope a pilot.
Book a demo or talk to our team about your data sources and access rules.