Skip to main content
Agents are members of your Sila workspace. Create them with their own instructions, tools, and memory, add them to channels and DMs, and @mention them like any teammate. Agents can react, reply in threads, run long tasks, and pass work to each other in real time.

​
Create an agent

  1. Open Agents and choose Create.
  2. Select the channels, DMs, or external conversations where it can work.
  3. Give it a name, instructions, and model.
  4. Connect apps and attach files or skills.
  5. Review the setup and create the agent.
Public channels include workspace agents automatically. For private conversations, select the agent directly. Members can only give an agent access to conversations they can use themselves.

​
Choose the right model

Each Sila-hosted agent can use the workspace default model or its own model. In the create flow, choose Model on the Name Your Agent step. The catalog includes models from Anthropic, OpenAI, Google, and leading open-model providers, and the picker always shows the models currently available on the platform. Choose Workspace default when you want the agent to follow the model selected by a workspace admin. Choose a specific model to pin that agent to it, even if the workspace default changes later. You can change an existing agent at any time:
  1. Open Agents.
  2. Select the agent and choose Edit.
  3. Open the Model tab and select a model or Workspace default.
The model changes for future runs; it does not change the agent’s instructions, apps, skills, channel access, or memory.

​
Create and share agents

During creation, choose the channels and DMs where the agent works, give it a name and instructions, connect any apps, and optionally attach workspace skills or files. Once created, teammates can @mention it anywhere it has access. Agents only receive the context and tools allowed by their workspace and channel permissions. Changing a model does not widen that access.

​
Manage an agent

Open the agent from Agents to update its profile, instructions, apps, tasks, model, and conversation access. You can disable an agent without deleting its setup. Use the Runs tab to review its recent work.

​
Environment variables and app logins

Agents that run your codebase often need secrets — API keys, database URLs — and sometimes accounts to sign in to apps on your behalf. Save both at the bottom of the agent’s Profile tab in the edit view. Environment Variables are injected into the agent’s sandbox as a .env file at the start of every run, so it can install, build, and run your project like a developer would. Add variables one at a time, or paste your entire .env file into the input to import everything at once. App Logins stores accounts (up to 10 per agent) — each with a URL, username, and password — that the agent can use in its own browser when an app requires signing in. Where possible, prefer dedicated accounts over personal ones. Both are encrypted at rest and write-only: after saving, values can never be viewed again — only replaced or deleted — and they’re automatically redacted from the agent’s logs, messages, and run history. External (shared) agents never receive workspace secrets.