Docs

Build and govern your agent’s brain.

Two ways in: the dashboard for a guided flow, or your IDE over MCP for full control. Both write to the same governed brain.

01 · Concept

The model

Every agent has a governed brain: a scoped vault of approved workflows, rules, and escalation conditions. The agent acts only on what a human on your team approved.

You put knowledge into that brain two ways, and both land the same place: as proposed memory that governs nothing until you approve it.

Dashboard: a guided flow. Add workflows, import a brain, review and approve. No code.
IDE over MCP: shape your agent from Cursor, Claude Code, or Codex. Full control for developers.
02 · Easy path

Dashboard

Sign up, create an agent, and its brain is provisioned. From the dashboard you add memory, import a starter brain as JSON, and review what’s pending, all governed.

Create an agent.
Add workflows directly, or Import JSON to seed a starter brain.
Review pending memory and approve what should govern.
Test the agent on a real ticket, then activate.
03 · Power path

IDE over MCP

Agents include IDE access. Connect your coding agent to your brain over MCP and manage it from the tools you already use.

MCP endpoint
# Connect Cursor, Claude Code, or Codex to your agent’s brain
url: https://mcp.conxt.dev
auth: your conxt key · from the dashboard
scope: this agent’s governed vault

Through MCP you use the same memory tools the dashboard calls (add, propose, approve), writing to the same governed brain. Available on Developer and Team plans.

04 · Format

Import format

Seed a brain in bulk with a JSON file of workflows. Each workflow needs a name and use_when conditions (in the vocabulary real tickets use) so the agent can match against it.

brain.json
{ "memories": [
  { "type": "workflow",
    "content": {
      "workflow": "Restore subscription entitlement",
      "use_when": ["paid for pro but shows free"],
      "steps": ["verify payment", "restore access"],
      "escalate_when": ["payment unconfirmed"]
    } }
] }

Imported memory lands as proposed: it governs nothing until you approve it in the dashboard.

05 · Governance

Governance

Whether knowledge enters through the dashboard or MCP, the rule holds: the agent proposes, a human approves, and only then does it govern. Rejected memory never governs anything, and every decision is audited.

Read the governance model →
06 · API

Agent API

Programmatic access to run classification and execution against your agent is rolling out with early access.

The agent execution API is in early access. Request access and we’ll onboard you directly.

For the memory layer beneath conxtagents, the full API is documented on conxt.dev.

conxt.dev memory docs →