Agents

Agents are autonomous AI assistants that run on dedicated VMs on Aleph Cloud. Each agent has its own system prompt, model, skills, and persistent workspace.

Creating agents

Use the five-step creation wizard (see Getting Started) to configure:

• Name — A memorable identifier (up to 50 characters)
• System prompt — Instructions that define your agent’s personality and behavior. You can write a custom prompt or select a template to pre-fill everything
• Skills — Optional capabilities (developer, productivity, web3 categories) that teach your agent specific workflows
• Model — The inference model (qwen3-coder-next or glm-4.7)
• Telegram bot (optional) — Link a Telegram bot token to let users chat with your agent via Telegram

Templates

Templates pre-fill the system prompt, model, and skills in one click. Available templates are fetched from the gallery and shown as pills in the Soul step. Selecting a template loads its full system prompt asynchronously. You can deselect a template by tapping it again, which resets to default values.

Deployment status

After creation, agents go through a deployment process on Aleph Cloud. The status cycle is:

Status	Meaning
pending	Queued for deployment
deploying	VM is being provisioned and configured
running	Agent is live and responding to health checks
failed	Deployment or health check failed

The agent detail page shows a deployment progress view with a circular progress indicator, step-by-step timeline, and a terminal log with real-time messages.

Health monitoring

Running agents are checked every 15 seconds via a health endpoint. The detail page shows one of three states:

• Healthy (green dot) — Agent is online and responding
• Unhealthy (orange warning) — VM is deployed but the agent service is down
• Failed (red error) — Deployment failed entirely

When an update is available (the agent’s version is behind the current version), an “Update available” banner appears showing the version difference.

Updating agents

There are two ways to push changes to a running agent:

• Update — Pushes new code and configuration to the existing VM. The agent restarts but keeps its workspace data. Faster (~30 seconds).
• Rebuild — Destroys the current VM and creates a new one from scratch. The agent is unavailable during redeployment (~2-3 minutes). Use this when the VM itself is broken.

Editing agents

Tap Update Configuration on the agent detail page to open the edit screen. The edit screen has four tabs:

• Soul — Edit the system prompt with a full-screen code editor. Shows line numbers, character count (max 8,000), and prompt complexity indicator. Templates can be applied here too.
• Skills — Toggle skills on or off, organized by category.
• Parameters — Change the agent name and model.
• Integrations — Enable/disable Telegram bot, update bot token, set owner Telegram ID, manage Telegram contacts.

After editing, you can either:

• Save — Saves configuration without redeploying (changes take effect on next redeploy)
• Save & Update — Saves and immediately pushes an update to the running VM

Deleting agents

On the agent detail page, scroll to the Management section and tap Delete Agent. A confirmation dialog appears. Deletion destroys the VM and removes all agent data permanently.

Telegram integration

Agents can be linked to a Telegram bot, allowing users to chat with the agent via Telegram:

1. Create a bot via @BotFather on Telegram
2. Enter the bot token during agent creation (Step 1) or in the edit screen (Integrations tab)
3. Optionally set your Telegram user ID to be auto-approved for chatting

The agent detail page shows Telegram bot status (connected/disconnected) and a link to manage Telegram contacts (approve, block, or remove users and groups).