Skip to content

using-codegen.md

Latest commit

 

History

History
261 lines (190 loc) · 5.64 KB

using-codegen.md

File metadata and controls

261 lines (190 loc) · 5.64 KB

Using CodeGen API with Forge

Forge supports CodeGen's agent-based code generation API as a backend for building projects.

Quick Start

1. Get CodeGen Credentials

Sign up at codegen.com and get:

  • API Token: From your account settings
  • Organization ID: From your organization dashboard

2. Set Environment Variables

export CODEGEN_API_KEY="your-api-key-here"
export CODEGEN_ORG_ID="your-org-id-here"

Or add to your .env file:

CODEGEN_API_KEY=your-api-key-here
CODEGEN_ORG_ID=your-org-id-here

3. Configure Forge

Create or update forge.yaml:

generator:
  backend: codegen_api
  api_key: ${CODEGEN_API_KEY}
  org_id: ${CODEGEN_ORG_ID}
  timeout: 600  # 10 minutes

Or use the global config at ~/.forge/config.yaml:

generator:
  backend: codegen_api
  api_key: ${CODEGEN_API_KEY}
  org_id: ${CODEGEN_ORG_ID}

4. Build Your Project

# Plan your project
forge chat

# Build with CodeGen
forge build --project your-project-id

Forge will automatically:

  1. Create CodeGen agent runs for each task
  2. Poll for completion
  3. Extract generated files
  4. Write them to your project directory

How It Works

Agent-Based Generation

CodeGen uses autonomous agents that:

  • Understand project context
  • Write production-ready code
  • Follow best practices
  • Generate tests automatically
  • Handle dependencies correctly

Parallel Execution

Forge leverages CodeGen's parallel execution:

  • Independent tasks run simultaneously
  • Up to 10 concurrent agent runs (respecting rate limits)
  • Faster overall project generation

Integration with KnowledgeForge

Forge enriches CodeGen prompts with:

  • Relevant patterns from the pattern library
  • Project context and tech stack
  • Dependency information
  • File structure expectations

This ensures CodeGen generates code that follows best practices and integrates seamlessly.

Example Session

# Initialize project
forge init my-api

# Start planning conversation
forge chat

> What would you like to build?
A REST API for task management with authentication

> What tech stack?
FastAPI, PostgreSQL, SQLAlchemy, JWT

> Any specific requirements?
- CRUD operations for tasks
- User authentication
- Task filtering by status
- Due date reminders

# (conversation continues...)

done

# Build with CodeGen
forge build --project my-api-20251207

# Output:
# ⚒ Forge Build System
#
# Project: Task Management API
# Stage: planning
#
# Found 12 tasks to build
# Backend: codegen_api
#
# Creating CodeGen agent runs...
# [1/12] Generating database models... ✓
# [2/12] Generating API endpoints... ✓
# ...
# [12/12] Generating tests... ✓
#
# ✓ Build completed in 8m 42s
# ✓ Generated 28 files

Advanced Configuration

Custom API Endpoint

If you're using a self-hosted CodeGen instance:

generator:
  backend: codegen_api
  api_key: ${CODEGEN_API_KEY}
  org_id: ${CODEGEN_ORG_ID}
  base_url: https://your-codegen-instance.com/v1

Timeout Settings

Adjust timeout for complex tasks:

generator:
  backend: codegen_api
  api_key: ${CODEGEN_API_KEY}
  org_id: ${CODEGEN_ORG_ID}
  timeout: 900  # 15 minutes for complex tasks

Repository Context

If your project is already in a CodeGen-connected repository:

generator:
  backend: codegen_api
  api_key: ${CODEGEN_API_KEY}
  org_id: ${CODEGEN_ORG_ID}
  repository_id: "your-repo-id"  # Optional: provides additional context

Rate Limits

CodeGen API has rate limits:

  • Agent creation: 10 requests per minute
  • Status checks: 60 requests per 30 seconds

Forge automatically:

  • Respects rate limits
  • Implements exponential backoff
  • Retries failed requests

Troubleshooting

"API key is required"

Make sure CODEGEN_API_KEY is set:

echo $CODEGEN_API_KEY
# Should output your API key

# If not set:
export CODEGEN_API_KEY="your-key"

"Organization ID required"

Set CODEGEN_ORG_ID:

export CODEGEN_ORG_ID="your-org-id"

"Rate limit exceeded"

Forge will automatically retry with exponential backoff. If you consistently hit limits:

  1. Reduce max_parallel in config
  2. Increase timeouts
  3. Contact CodeGen support for higher limits

Agent runs fail

Check the CodeGen dashboard at codegen.com to:

  • View agent run details
  • See error messages
  • Review generated code
  • Resume failed runs

No files generated

CodeGen's response format may vary. If Forge isn't extracting files correctly:

  1. Check .forge/logs/ for detailed output
  2. Manually review the CodeGen agent run
  3. Report the issue on GitHub

Comparison with Claude Code

Feature CodeGen API Claude Code
Execution Cloud-based agents Local CLI tool
Parallel tasks Yes (10/min) Yes (limited by machine)
Cost Pay per agent run API costs only
Setup API key only CLI installation
Web dashboard Yes No
Offline No Yes (with local model)

Best Practices

  1. Start small: Test with 2-3 tasks before full project
  2. Review output: Check CodeGen dashboard for quality
  3. Iterate: Use forge iterate to refine generated code
  4. Save sessions: CodeGen keeps history in dashboard
  5. Monitor costs: Track agent runs to manage expenses

Next Steps

Questions? Open an issue on GitHub