DevFlow AI

Never lose your AI context when token limits hit.

DevFlow has two modes—use one or both:

Use Case	Setup	How
Context recovery (Cursor, Copilot, Claude Code, any AI)	Zero	Run npx github:abhishek5878/devflow → paste into Claude.ai (copies to clipboard)
Proxy routing (Continue.dev, Roo Code, Cline)	Add API key, start proxy	Auto-failover across Claude, GPT-4, Gemini when one hits limit

See INSTALL.md for step-by-step setup. Extension: Command Palette → DevFlow: Get Started.

Landing page: landing/index.html — deploy to Vercel, Netlify, or GitHub Pages. Preview: npm run landing:serve (http://localhost:3001).

---

Phase 1: Local Proxy Router

OpenAI-compatible endpoint at http://localhost:8080/v1 that:

• Routes requests across multiple AI providers (Claude, GPT-4, Gemini)
• Tracks token usage with exact accuracy
• Silently fails over on HTTP 429 (rate limit) without interrupting the developer

Quick Start

# Install
npm install

# Configure (optional - copy .env.example to .env)
cp .env.example .env
# Add OPENAI_KEY, ANTHROPIC_KEY, etc.

# Run
npm run dev

IDE Configuration

Point any OpenAI-compatible client at DevFlow:

{
  "ai.endpoint": "http://localhost:8080/v1",
  "ai.apiKey": "devflow-local"
}

Works with: Continue.dev, Roo Code, Cline, and any tool accepting custom OpenAI endpoints. OpenClaw — DevFlow plugin adds devflow_context_snapshot for AI handoff from the agent.

Test Failover (No API Keys Required)

# Terminal 1: Start with mock providers
DEVFLOW_MOCK_LIMITS=1 npm run dev

# Terminal 2: Run limit exhaustion test
npm run test:limits

You'll see Claude "exhaust" after ~2-3 requests, then silent failover to GPT-4.

Full Test Suite

npm test          # Build, context, extension
npm run test:all  # Above + proxy failover + client (starts proxy automatically)
npm run test:userflow  # End-to-end user flow (Context Snapshot + Proxy)

For manual proxy test:

DEVFLOW_MOCK_LIMITS=1 npm run dev   # terminal 1
npm run test:proxy                  # terminal 2

Basic Connectivity Test

npm run dev    # terminal 1
npm run test:client   # terminal 2 (requires OPENAI_KEY or ANTHROPIC_KEY)

API Endpoints

Method	Path	Description
POST	/v1/chat/completions	OpenAI-compatible chat completions (proxied)
GET	/v1/models	List available models
GET	/v1/health	Provider status & token usage

Provider Priority

1. Claude (100k limit)
2. GPT-4 (80k limit)
3. Gemini (90k limit)
4. Ollama (∞, local — run ollama run llama3.2)
5. Amazon Q, Codeium (∞, when integrated)

At 90% capacity: warning logged. At 100%: automatic failover. Ollama used as free fallback if running.

OpenClaw (optional provider)

If you run OpenClaw locally, DevFlow can route through it as another provider.

1. Enable the OpenAI-compatible endpoint in OpenClaw (required):

Add this to ~/.openclaw/openclaw.json (or via the Control UI Config tab), then restart with openclaw gateway restart:

{
  "gateway": {
    "http": {
      "endpoints": {
        "chatCompletions": {
          "enabled": true
        }
      }
    }
  }
}

2. Point DevFlow at OpenClaw (optional overrides, defaults shown):

# .env
DEVFLOW_OPENCLAW_URL=http://localhost:18789/v1
DEVFLOW_OPENCLAW_MODEL=openclaw-default

When these are set and OpenClaw is running, DevFlow will treat it like any other provider and fall back to it after Claude/GPT/Gemini, before Ollama.

---

Phase 2: Context Engine

Recovery tool for closed systems (Cursor, GitHub Copilot). Generate context.md for seamless handoff to any AI.

npx github:abhishek5878/devflow

Default: Copies to clipboard. Paste into Claude.ai. Done.

Options: --no-copy, --skip-tsc, --target claude|cursor|continue (paste hint), --output path

Output: Git history, stack detection, project rules, "For the Next AI" handoff. One command. Zero friction.

---

Phase 3: VS Code Extension

From repo root (/Users/abhishekvyas/devflow):

npm run ext:setup

Or step by step:

npm run build
cd devflow-vscode && npm install && npm run compile

Then open the devflow-vscode folder in VS Code and press F5.

• Cmd+Shift+D (Mac) / Ctrl+Shift+D (Win/Linux) — Generate context.md
• Status bar — Proxy status and token usage (click for dashboard)
• Commands — Start Proxy, Stop Proxy, View Dashboard

---

---

Installation (from .vsix)

See INSTALL.md for Quick Start, configuration, and troubleshooting.

Package the extension (requires Node 20+):

cd devflow-vscode && npm install && npm run package
# Creates devflow-0.1.0.vsix

On Node 18, use the extension from source with F5 (see Phase 3 above).

---

Roadmap

• Phase 4: Publish to VS Code Marketplace — see PUBLISH.md for steps