Bring Your Own Agent (BYOA)
Use your preferred AI coding agent — Claude Code, Cursor, GitHub Copilot, or Antigravity — to complete technical assessments in your own environment.
Overview
BYOA lets you take coding assessments using the AI tools you already know. Instead of working in a browser-based IDE, you solve problems locally in your own editor with your own AI agent. InterviewLM captures your work through a lightweight CLI tool — no sandbox, no restrictions on your environment.
How It Works
A lightweight CLI sets up your workspace with the problem, starter code, and agent-specific configuration. As you work, it captures your AI interactions and code changes. The same evaluation pipeline scores your submission.
Choose Agent
Run Setup
Solve Problem
Submit
Supported Agents
Choose the agent you're most comfortable with. All agents capture your work for evaluation — agents with MCP support provide richer interaction data.
Claude Code
Config: CLAUDE.md + .mcp.json
Anthropic's CLI agent for your terminal. Rich MCP integration captures every prompt and response.
Cursor
Config: .cursorrules + .cursor/mcp.json
AI-powered code editor. MCP integration captures interactions through Cursor's built-in protocol.
GitHub Copilot
Config: .github/copilot-instructions.md
GitHub's AI pair programmer in VS Code. File watcher captures code evolution and diffs.
Antigravity
Config: Knowledge base config
Google's agentic IDE. File watcher provides universal capture of your work.
Choose Your Agent
When you open your assessment link, you'll see the agent selection screen. Pick "Browser IDE" for the traditional in-browser experience, or choose your preferred AI agent for a local BYOA session.
Choose Your Environment
Browser IDE
DefaultClaude Code
Cursor
Copilot
Org-controlled availability. Your employer decides which agents are available for each assessment. You'll only see agents that have been enabled for your specific assessment.
Run the Setup Command
After clicking "Start Assessment," you'll see a setup command. Copy it and run it in your terminal. The CLI downloads the problem, creates your workspace, and connects to InterviewLM.
Waiting for Connection
Run the setup command below to connect your agent.
Setup Command
npx @interviewlm/cli init abc123def456...$ npx @interviewlm/cli init abc123def456...
InterviewLM CLI v1.0.0
Downloading problem and starter code...
Creating workspace: ~/ilm-assessment/
Writing agent config (CLAUDE.md, .mcp.json)...
Starting MCP server on port 3456...
Starting file watcher...
Ready! Open ~/ilm-assessment/ in Claude Code
Tip: Run "cd ~/ilm-assessment && claude" to start
Workspace created
Problem statement, starter files, test files, and agent config all in one directory
MCP server running
Captures rich per-turn data from agents that support Model Context Protocol
File watcher active
Captures code diffs every 30 seconds as a universal fallback
Solve the Problem
Open the workspace in your chosen agent and start coding. Work naturally — ask your AI for help, run tests, iterate. Everything is captured automatically.
What Gets Captured
- Your prompts to the AI agent
- AI responses and tool calls
- File changes and code diffs
- Test runs and results
What Gets Evaluated
- Problem-solving approach
- Quality of AI collaboration
- Code quality and correctness
- Test pass rate
AI collaboration is encouraged. We evaluate how effectively you work with AI, not whether you can solve problems without it. Use your agent naturally.
Submit Your Work
When you're done, submit from the CLI. This sends a final snapshot of all files and triggers evaluation.
$ npx @interviewlm/cli submit
Collecting final workspace snapshot...
Uploading 12 files (4 modified, 2 created)...
Sending submission to InterviewLM...
Assessment submitted successfully.
Session summary:
Agent turns: 14
File edits: 23
Duration: 47 minutes
Your submission is now being evaluated. Results will be shared by the hiring team.
You can also submit from the web dashboard as a fallback — look for the "Submit Assessment" button on the monitoring page.
Monitoring Dashboard
While you work, the web page stays open as a real-time dashboard showing your connection status, time remaining, and activity stats.
Claude Code Connected
Your agent is connected and recording activity.
42:15
8
15
34
Prefer to submit from your agent's CLI. Use this only as a fallback.
Frequently Asked Questions
Do I need to install anything?
You need Node.js (v18+) to run the CLI via npx. The CLI itself requires no global installation — npx handles it automatically. You also need your preferred AI agent installed (Claude Code, Cursor, etc.).
What if BYOA isn't shown on my assessment?
BYOA availability is controlled by the employer. If you only see "Browser IDE" as an option, that means the employer has restricted this assessment to the browser-based environment.
Can I switch between browser and BYOA after starting?
No. Once you start the assessment with a delivery mode, it's locked in for that session. Choose carefully before clicking "Start Assessment."
What if my internet drops during a BYOA session?
The CLI queues events locally and retries when connection is restored. Your local work is never lost. The dashboard will show a disconnected state until the CLI reconnects.
Is BYOA evaluated differently from the browser assessment?
No. All BYOA events are normalized into the same format as browser sessions. The evaluation pipeline scores both identically. The only difference is that your evaluation report will show which agent you used.
Can I use multiple AI tools simultaneously?
You should use the agent you selected during setup. The CLI is configured for that specific agent. Using additional external AI tools outside the tracked agent is not captured and may be flagged as an integrity concern.