Integrations
How the Company Brain connects to your agents: a central backend for memory and data, plus a lightweight plugin that gives each agent access without code changes.
Integrations are how the Company Brain embeds into your company and reaches every agent. The design has two parts that work together: a central backend that holds everything, and a local plugin that lives inside each agent and connects it to the company brain.
The point is that agents don't have to be rebuilt to use the brain. The backend is where memory and data live; the plugin connects each agent to it, so every agent gains shared memory and company knowledge once the plugin is installed.
The two parts#
One backend manages everything: the company's shared memory, the connected data platform, and the Ontology Graph. It is where agents read, write, and where access is governed. All agents, wherever they run, draw from this one brain.
A lightweight module that installs into each agent across the company. It connects the agent to the central backend and handles all the work of using memory and company data on the agent's behalf, with no changes to the agent itself.
How the plugin works#
The plugin hooks into the agent's input, so using the brain is automatic rather than something each agent has to be programmed for:
Every time the agent receives a prompt, the plugin steps in first, before the agent acts.
It pulls in the relevant shared memory and company knowledge from the central backend and adds it to the agent's context, so the agent answers grounded in what your company actually knows.
Afterward, it records what's worth keeping back into shared memory, so the context the agent gained is available to every other agent next time.
The agent never has to call an API explicitly. The plugin does the retrieval, the grounding, and the write-back transparently on every turn, all governed by the company's access policies.
Works with any agent#
Because the plugin does the work, the same integration can be added to each agent in the same way:
The plugin installs into agents such as Codex, Claude Code, Openclaw, Cursor, and Hermes, each backed by shared memory and company knowledge.
Connect a local or open-source model, for example a sandboxed agent that never leaves your perimeter. The same plugin connects it to the brain, even when the model stays entirely inside your network.
Point an existing internal agent platform at the central backend so its agents share the same governed brain instead of siloed per-team stores.
Where you'd rather call the brain directly, applications can query it for grounded answers while streaming reasoning, rendering charts and visuals, and showing citations and the confidence signal.
One control plane for everything#
This two-part design is what delivers a single control plane:
- One place for connections: sources are connected once at the backend; no agent maintains its own integrations.
- One place for memory: every agent reads and writes the same shared memory through the plugin.
- One place for governance: access, permissions, and audit are defined and enforced centrally, then applied to every agent. See Security.
The alternative, N agents × M sources of bespoke integrations and N separate memory stores, collapses into one backend and one plugin. New sources and new knowledge become available to every agent at once.
Enroll devices and deploy the plugin across your company with Team Sync.