Create Your First Pylon Agent

What you’ll accomplish

You’ll create an autonomous Agent in Pylon, try it with a real run, and publish it - starting from a template or a one-line description.

What you’ll need

• Access to Pylon (see Availability)
• At least one chat model available to your account - the editor will tell you if none are configured yet
• A clear idea of the job, one sentence is enough

Steps

1. Open Agents and start a new one

In the left navigation under Build, select Agents, then Create agent. You’ll land on the New agent page: “Start from a template or describe what you want, then refine.”

2. Seed it - describe or pick a template

Both options are optional; they pre-fill the tabs below, and you can skip straight to filling them in yourself.

• Describe your agent - type one line (the placeholder shows the shape: “Summarizes new GitHub PRs and posts a digest to Slack.”) and select Generate. Pylon turns the description into a starting configuration.
• Or start from a template - pick from the gallery: Blank agent, Code Reviewer, QA Agent, Support Triage, PR Backup (GitHub), or Research Assistant.

You’ll see the editor tabs below populate with the seeded configuration.

3. Fill in the Overview tab

Set the Name, an optional Description, choose the Model (the dropdown lists your account’s available chat models), and write the System prompt - the standing instructions that shape how the agent works.

4. Give it tools and skills

• Tools tab - the built-in tools are always available (they’re listed read-only under “Built-in tools - always available”). To add more, attach an MCP server: pick an existing one or select Register a server and give it a name.
• Skills tab - toggle on any Skill catalog entries the agent should know. The agent loads their instructions before acting.

5. Create it, then run it

Select Create agent. You’ll see the toast “Agent created” and land on the agent’s page, with a Draft badge and Save, Run, and Publish buttons in the header.

Select Run to try it - you’ll see “Run started” with your agent’s name. Watch progress under Monitor → Activity while per-agent run history is still rolling out.

6. Publish when you’re happy

Select Publish. The badge flips from Draft to Active, and the published revision is what triggers and other users get. Keep iterating with Save + Run; publish again to promote changes.

If it didn’t work

• The Create button is disabled: “No chat models are available for your account yet.” Your administrator needs to enable a model for your account in Canopy first.
• “Couldn’t turn that description into a config.” The Generate call failed; try again, simplify the description, or fill the tabs directly.
• “You have reached your agents plan limit.” Your plan caps the number of Agents; archive one or upgrade.

Next steps

• Schedule the agent - run it on a cadence
• Agent configuration reference - Limits, Sandbox, Config JSON, and every other tab