- Olbrain
- Olbrain is The Agent OS for building, running and governing agent systems that work together. Specialized systems connect through relevant shared context, clear responsibilities and handoffs, business-wide goals and common governance. Connections require scoped engineering and validation.
- The orchestrator
- The lead orchestrator inside Olbrain: an AI agent that builds the agent systems, directing a team of specialist sub-agents through discovery, build, and deployment.
- Agency Protocol
- Olbrain’s mechanism for giving each agent a self-identity — one accountable, exclusive actor — so its decisions are attributable and the enterprise can answer for them.
- Self-identity
- The agent’s own self: who it is and that it remains the same actor across time, answerable for its record. Distinct from external identity — authorization and access management — which governs only what an agent may access.
- Agent brain
- The definition of an agent — identity, tools, knowledge, channels, guardrails — that compiles into a deployable system prompt and runtime configuration. Versioned and immutable on publish.
- BRD
- A business-requirements document produced during discovery; the agent brain is generated from it.
- Workflow
- A directed graph built from typed step executors that an agent runs; steps fan out per item and aggregate. Olbrain ships 37+ typed step executors.
- Tool · MCP
- A capability an agent calls to act in the world, implemented as a Model Context Protocol (MCP) tool server.
- Connector
- An integration into an enterprise system or API an agent uses to access data and take action.
- Knowledge
- Enterprise data — documents and systems — that grounds an agent’s answers and actions.
- Trigger
- An event that starts an agent or workflow: a conversation, another agent, an external event, or a schedule.
- Runtime
- The execution mode for an agent. Three are live: conversational, workflow, and research.
- Agent types
- The five kinds of agent: Concierge (conversational), Operator (workflow), Analyst (research), Orchestrator (coordination), Designate (role-holding — humans supervise). See capabilities.
- Context Service
- Olbrain’s persistent, judge-gated agent memory — append-only, scoped global → agent → session.
- CS-packet
- Compiled working context, including memory and history, supplied to a model call. A context carrier, neither an identity construct nor a billing unit.
- Audit log · receipt
- An immutable, append-only record of every agent action; each action emits a signed, PII-free receipt.
- Audit packet
- An on-demand record of an agent action — produced by replaying it — for InfoSec, Risk, or a regulator.
- CoF · Core Objective Function
- An agent’s declared purpose, bound at creation, that orients its behavior and stays stable over time — a first-class primitive.
Start building with Alchemist