Bring an agent to production
An engineer or a process owner has built an agent that works. Now more people need to use it, it has to reach company systems, and someone has to be able to check what it does. This page describes the path with TeraBrain NX.
Credentials kept out of the code, least privilege on every port, access decided channel by channel.
Roles, groups, versions with an author and an activity log, all in one place.
You start from ready-wired patterns (recipes) instead of from scratch.
From prototype to production
An agent that started as a prototype usually works for a single person, with their own credentials, and no record of what it does. Five things change when it goes into production.
| Aspect | Prototype | With TeraBrain NX |
|---|---|---|
| Who uses it | Whoever built it | Organization members, groups of members or the public, depending on the chosen channel |
| Credentials | In the code or personal | Organization secrets referenced by name, or OAuth connections |
| Connection to systems | One-off solutions | TypeScript connectors, MCP servers, databases, customer machines |
| Traceability | None | A record of conversations, tool calls and changes |
| Changes | Invisible | Every change is a version with an author, and you can roll back |
The path in five steps
- Bring the agent into the projectIn TeraBrain NX an agent is a node with instructions, a model and tools. You bring over the instructions (in Markdown) and choose the model. See How to create a new agent.
- Connect your systemsDatabases, files, TypeScript functions, MCP servers or on-premise machines become the agent's tools. See Code and integrations.
- Choose the channelWeb app, Telegram, mailbox, phone, scheduler or MCP server: each one brings conversations to the agent and has its own access control. See Channels.
- Decide who can do whatRoles, groups and least privilege on ports. See Security.
- Test, publish, observeChecks and Try it before saving; afterwards, the logs show every conversation and every call. See Governance and audit.
- Instructions
- Local tests
- One person
- Agent
- Tools
- Channels
- Access
- Logs
- Colleagues
- Business systems
- Control owners
Three profiles, three routes
- Process owner or manager. Starts from a recipe, or describes what they want in natural language to the integrated Builder, which builds the project and says what is left to do (for example setting a secret).
- Engineer or developer. Connects an external assistant (Claude Desktop, Cowork, Claude Code, Codex) to their profile with a personal token, and works with it on the project. Their own code lives in TypeScript connectors with a git repository per node. See Building with external assistants.
- Software house or system integrator. Uses connectors and on-premise connectors to reach customers' systems, web apps with a built-in assistant for software already installed, and recipes as a starting point.
What to know before you start
Related pages
How to create a new agent
Step by step: project, agent node, instructions, model, tools, channel, testing and publishing.
Channels: how people reach agents
Web app, Telegram, mailbox, phone, scheduler and MCP server: how they connect and how you control who can use them.
Security
Roles, groups, credentials, enterprise sign-in and least privilege: how to protect an agent system.
Building with external assistants
Using Claude, Codex or other assistants to build and develop TeraBrain NX projects, with the person's own rights.
Have an agent to take to production?Letβs talk.
We show you TeraBrain NX on your case: systems to connect, access and controls.
