Overview
An agentic QA engine that finds bugs, writes tests, reviews code, and opens pull requests with fixes.
The Paragon Agent is an agentic QA engine that clones your repository, analyzes your code for issues, writes tests, and opens pull requests with fixes. Use it from the dashboard, trigger it from GitHub comments and Slack, or automate it on schedules and events.
What It Can Do
- Find and fix bugs — Point it at an issue, error, or flaky test and it investigates and patches
- Write and improve tests — Generate unit, integration, and E2E tests and increase coverage across your codebase
- Review code quality — Catch security vulnerabilities, performance issues, and regressions before they ship
- Validate changes — Run builds, execute test suites, and verify nothing breaks
- Browse and interact with web apps — Use a real browser in a sandbox to test UI, fill forms, and validate behavior
How It Works
- Select a repository — Choose a connected GitHub repository and branch
- Describe the task — Tell the agent what you want reviewed, tested, or fixed
- Agent executes — The agent clones your repo in an isolated sandbox, analyzes the code, runs tests, and makes changes
- Review the PR — Review the diff in the dashboard or on the generated pull request
Session Modes
| Mode | Best For | Duration | How It Works |
|---|---|---|---|
| Standard | Quick bug fixes, test generation, code review | Minutes | Single agent processes your prompt directly |
| Grind | Comprehensive test coverage, large-scale refactors, multi-file fixes | Hours | Plans first, then delegates to parallel workers (LRA) |
Grind mode is powered by the Long Running Agent (LRA) — a 2-phase orchestrator that plans, gets your approval, then executes with parallel workers.
Models
| Model | Speed | PCU Cost | Best For |
|---|---|---|---|
| Max | Slowest | Highest | Complex debugging, nuanced code review, architecture-level issues |
| Mid | Balanced | Moderate | Most tasks — good balance of quality and speed |
| Fast | Fastest | Lowest | Simple fixes, test generation, straightforward patches |
MCP Integrations
The agent can connect to external services via MCP (Model Context Protocol) to expand its capabilities:
- Linear — Read and update issues
- Slack — Send messages and read channels
- Jira — Manage tickets
- Notion — Access documentation
- Sentry — Investigate errors
- Vercel — Check deployments
- Supabase — Query databases
- Cloudflare — Manage workers and DNS
- AWS — Interact with AWS services
- MongoDB — Query collections
- Custom MCP servers — Connect any MCP-compatible service via stdio, HTTP, or SSE
Configure MCP integrations from the agent settings panel.
Usage Analytics
The Usage panel in the sidebar shows:
- PR metrics — PRs created, merged, merge rate, PCUs per merged PR
- Session metrics — Total sessions, tokens used, total cost, average duration
- Pull request chart — Status breakdown (merged, open, closed) over time
- Model usage breakdown — Sessions, tokens, and cost per model tier
- Time range — Daily or weekly aggregation over the last 90 days
Next Steps
Sessions
Create and interact with agent sessions
Grind Mode
Tackle complex tasks with parallel agents
Automations
Trigger the agent on schedules and events
Integrations
Connect the agent with GitHub and Slack