Skip to main content

Quick Start

This guide validates a new installation by configuring one model, creating a workspace, and running the first session and ticket.

0. Sign in​

Open the HTTPS URL configured by your operator and sign in with Google, Microsoft, or Generic OpenID Connect. The first successfully registered user becomes Admin; complete this bootstrap with the intended administrator account before opening access to other users.

Access tokens are short-lived; refresh tokens last 7 days. Auth state stays in sync across browser tabs automatically.

1. Pick a model​

The built-in Polygent Code agent requires no separate provider selection. In Settings, configure credentials for one approved model, make that model visible, and set it as the default. Local or subscription-backed models also require their approved runtime and login on every eligible session host.

Tip: Choose a model from a single flat dropdown, switchable per session, per workflow, per ticket, or per bot — there is no provider to select. You supply your own model API keys (BYOK).

2. Create a workspace​

A workspace binds Polygent to a Git repository and is the unit of multi-tenant isolation.

  1. Click New Workspace in the sidebar
  2. Fill in:
    • Name and Default branch (e.g. main)
    • Git URL + Git PAT (GitHub or Azure DevOps)
    • Ticket source (None, GitHub, or TFS) if you want external sync
  3. Click Create — you become the first workspace user automatically

3. Run your first session​

A session is an AI run inside an isolated Git worktree, allowing concurrent sessions without sharing one working directory.

  1. From My Work in the sidebar, click the plus button
  2. Pick a mode:
    • Develop — ephemeral worktree, task-focused (closes on completion)
    • Chat — persistent worktree, ongoing conversation
  3. Pick a model
  4. For Develop mode, optionally enable Edge Cases, Code Review, Verification, Update Tests, or Update Docs
  5. Type your first message — use @ to mention files and / to call slash commands or skills

You'll see live updates as they happen: agent messages, tool calls, file changes, and token/cost tracking.

4. Run sessions in parallel​

Click the plus button again and start another session. Each session gets its own isolated git worktree, so multiple features run concurrently without collisions on disk. The sidebar shows all active sessions and tickets with live progress, broadcast to every connected browser — updates appear automatically, with no manual refresh.

5. Create a ticket (PM/manager flow)​

Tickets turn AI sessions into trackable, queueable work.

  1. From My Work in the sidebar, click the plus button and choose Create Ticket
  2. Fill in title, description, and (optionally) attach files
  3. Choose:
    • Workflow — multi-step procedure to run
    • Model
    • Developer + QA assignees
    • High priority (jumps the queue)
  4. Click Queue — the ticket waits until an eligible session host has ticket capacity

Tip: Use Request bot and ask him to create a ticket for you — he can fill in all the details from a plain-language description.

Tickets flow through stages: Pending → Queued → Implementation → Developer Approval → QA Approval → Pull Request → Completed (a ticket can also end in Failed, Canceled, or Merge Conflicts). The Activity Feed and Rework History panel show the timeline, including QA rejection rounds.

Tip: Use Start Ticket Templates to capture preset configurations for common ticket types.

6. Plan a feature in plain language​

Use the AI Planner when someone describes a feature in plain language without the implementation details. Plain language goes in, a structured spec comes out.

  1. From My Work in the sidebar, click the plus button and choose Plan
  2. Pick the workspace, model, and target branch
  3. Write a plain-language description and attach any context files
  4. Choose an auto-answer mode (MVP, Balanced, Features-Rich, or Custom) to let AI answer its own questions
  5. Step through the 6-step wizard: Configuration → Understanding → Clarifications → Recommendations → Specifications → Review
  6. Edit the generated plan on Review step, then click Create Ticket to convert it into a ticket and kick off implementation immediately

7. Review and merge​

Polygent runs three approval gates — Developer → QA → PR — with rejection feedback and file attachments. QA does not have to wait for a PR: with Deploy to Slot, an in-progress session is reachable from a live URL, so QA can test as the agent builds.

When a session or ticket reaches Developer Approval or QA Approval:

  1. Open the Tickets page and click into the ticket
  2. Approve, or reject with feedback (text and/or file attachments)
  3. Once merged, Polygent creates a PR (GitHub or Azure DevOps) or performs a direct merge, depending on the workspace's Merge Mode
  4. PR status (Open / Merged / Closed / Conflicts / Failed to Create) is tracked live

8. Try a bot​

Bots are workspace-scoped (or global) AI assistants with a fixed system prompt — great for fast Q&A and triage.

From My Work in the sidebar, click the plus button and choose your bot (e.g. Request for ticket creation, Debugger for debugging help).

9. Open a session in your IDE (optional)​

When you want to drop into a session locally:

  1. Open the Session page and click Open in IDE
  2. Copy the Bash or PowerShell one-liner and run it in your terminal
  3. The script downloads a single-use changes zip and unpacks it on top of your local clone
  4. Push uncommitted edits back with the matching Upload from IDE one-liner, or git push to your own remote and use Pull from Remote in the session

Tokens are minted with a 5-minute TTL and are single-use.

10. Schedule recurring work​

Use Automations to run sessions on a schedule (cron), once at a specific time, or in a continuous loop after each session completes — with missed-run recovery so scheduled work survives downtime. Pick the session type (Develop, Chat, or Bot), an inline prompt or a workflow, the model, and the trigger. Recurring and Loop automations auto-disable after 3 consecutive failures (One-Time on the first failure); Manual and HTTP Trigger automations are never auto-disabled.

11. Capture team knowledge​

Polygent has two systems that turn each session into reusable team knowledge:

  • Insights — auto-extracted from completed Develop sessions. Categorize, dismiss, or convert into tickets.
  • Memory Stores — per-workspace named buckets that agents read and write at runtime via MCP tools.

12. Run a Round Table​

When a decision needs multiple viewpoints:

  1. From My Work in the sidebar, click the plus button and choose Round Table
  2. Pick personas (Moderator, Devil's Advocate, Steelman, Senior Developer, QA Lead, Product Owner, Designer, or custom)
  3. Send a message and review the discussion from each selected perspective
  4. Call a vote, generate a running summary, or export the conversation to TXT or Markdown
  5. Convert the discussion into a ticket or plan with one click

What's next​