GitHub Integration
Plug AVCodex into the repos where your control system code, custom Q-SYS plugins, and SIMPL# libraries live. @mention your AVCodex agent in a PR, issue, or discussion and get a review that knows your Crestron modules, your Q-SYS Lua patterns, and the project's commissioning history.
Deploy your AVCodex agent as a GitHub App. Programmers @mention your bot in issues, pull requests, and discussions to get AI responses with full repository context. Built for control system code review, integration triage, security audits on remote-access modules, and keeping a programmer's Q-SYS Lua module library well-documented.
Note: GitHub integration requires a Builder plan or higher. <a href="https://app.avcodex.com/plans" target="_blank" rel="noopener noreferrer" > Upgrade to Builder </a>
| Capability | Description |
|---|---|
| Read repository files | Access SIMPL# source, Q-SYS Lua modules, configs, and AS-built notes |
| Review pull requests | Analyze diffs and changed files with project context |
| Search code | Find patterns, function names, or references to a specific Crestron device class |
| Triage issues | Categorize, label, and respond to new field issues |
| Post comments | Respond in issues, PRs, and discussions |
| Add labels | Auto-label by manufacturer, room, or severity |
GitHub uses a streamlined one-click manifest flow. No manual OAuth configuration needed.
1. Deploy to GitHub
Go to your agent's Share page > Deploy tab > Deploy on GitHub card. Click Add Deployment.
You'll be redirected to GitHub to create a GitHub App from a pre-configured manifest. Click Create GitHub App and confirm.
2. Install to Repositories
After creation, GitHub provides an install link. Select which repositories should have access. Common picks: your Crestron SIMPL# library, your Q-SYS plugin repo, the AS-built docs repo for a multi-room corporate refresh.
3. Start Using
Your bot is now active. @mention its slug in any issue, PR, or discussion to get a response. The slug is displayed in the integration status panel.
`` @acme-av-avcodex can you review this SIMPL# diff for null-reference risk? ``
When someone @mentions your bot:
- GitHub sends a webhook to AVCodex with the comment context
- Your bot reads the issue/PR details, including title, description, labels, and diffs
- Repository tools are registered so the bot can read files, search code, and review changes
- The AI generates a response using your agent's system prompt and knowledge base
- The response is posted as a GitHub comment in Markdown format
The bot shows a thinking indicator (eyes emoji) while processing and a rocket emoji when done.
| Event | Trigger | Bot Responds In |
|---|---|---|
| Issue comment | @mention in an issue | Same issue thread |
| PR comment | @mention on a pull request | Same PR thread |
| PR review comment | @mention on an inline code comment | Same review thread |
| Discussion comment | @mention in a GitHub Discussion | Same discussion |
| New issue | Issue opened (if auto-respond enabled) | New comment on the issue |
The bot only responds to human-created events. Comments from other bots are ignored to prevent loops.
When your bot responds, it has access to these tools for understanding your codebase:
| Tool | What It Does |
|---|---|
| Get file contents | Read any file by path (SIMPL# modules, Q-SYS Lua, configs, READMEs) |
| Get PR diff | View the complete unified diff of a pull request |
| Get PR files | List changed files with addition/deletion counts |
| Search code | Find patterns and references across the repository |
| List issues | Search and filter issues by state, labels, or assignee |
| Get comment thread | Read the full conversation history |
| Create comment | Post follow-up comments for multi-step responses |
| Add label | Apply labels for automated triage (crestron, q-sys, field-bug, room-3) |
| Add reaction | React to comments with emoji |
Auto-Respond to New Issues#
By default, the bot only responds when @mentioned. You can enable auto-respond to have the bot automatically comment on every newly opened issue. Useful for support repos where field techs file bugs against shipped control system code.
Toggle this in your agent's GitHub settings under Share & Deploy.
AVCodex provides GitHub-optimized templates to get started quickly:
| Template | Purpose |
|---|---|
| Code Reviewer | In-depth PR review tuned for control system code (SIMPL#, Q-SYS Lua, Python on Crestron 4-Series) |
| Security Reviewer | OWASP/CWE-focused audits, with extra focus on remote-access modules and credential handling |
| Issue Triage | Categorize field issues by manufacturer, room, severity |
| Docs Assistant | Explain a Q-SYS module library or generate AS-built notes from code |
| Release Notes Generator | Summarize merged PRs into release notes for a control system version |
Each template comes with an optimized system prompt and suggested starting messages.
The GitHub App requests these permissions during setup:
| Permission | Access Level | Why |
|---|---|---|
| Contents | Read | Read repository files and code |
| Issues | Write | Post comments, add labels |
| Pull Requests | Write | Post comments, review diffs |
| Discussions | Write | Post in GitHub Discussions |
| Metadata | Read | Access repo names and public metadata |
The bot is read-only for code. It can post comments and labels but cannot modify code or merge pull requests. No risk of an AI pushing untested SIMPL# to production.
| Resource | Limit |
|---|---|
| GitHub API requests | 5,000/hour per installation |
| Code search | 10 requests/minute |
| Comment length | 65,536 characters |
For typical usage (dozens of daily interactions across a programming team), you won't hit these limits. Code search is the tightest constraint, the bot uses it sparingly in high-volume scenarios.
GitHub integration uses your standard LLM token billing. There's no separate per-message fee for GitHub interactions. GitHub API calls (reading files, searching code) are free and not metered.
Bot not responding?
- Make sure you @mentioned the exact bot slug (shown in your GitHub settings)
- Verify the bot is installed on the repository in question
- Check that the webhook URL is active in your GitHub App settings
Responses seem generic?
- Add knowledge sources to your agent (Crestron device class docs, Q-SYS Lua API references, your team's coding standards)
- Customize the system prompt to focus on your use case (PR review, triage, AS-built docs)
- Use one of the pre-built templates as a starting point
Rate limit errors?
- Code search is limited to 10/minute. If your bot searches frequently, space out requests
- Consider using file reads instead of code search for known file paths (e.g., a specific module in your library)
Note: MCP server access requires a Builder plan or higher. See the MCP Setup Guide to get started.
*AVCodex · Your AV expertise. Amplified by AI.*