Context-aware AI coding assistant powered by DigitalOcean Gradient
Paste your GitHub repo. Ask anything. Get code that actually fits your codebase.
- Docker Desktop installed and running
- DigitalOcean Personal Access Token → cloud.digitalocean.com/account/api/tokens
- DigitalOcean Gradient Model Access Key → cloud.digitalocean.com/gradient → Serverless Inference → Model Access Keys → Create Key
- (Optional) GitHub Token → github.com/settings/tokens — increases rate limit from 60 to 5000 requests/hr
git clone https://github.com/devu729/devmate-v2
cd devmate-v2cp .env.example .envOpen .env and fill in your 3 keys:
DO_GRADIENT_API_KEY=dop_v1_xxxxxxxxxxxxxxxx
GRADIENT_MODEL_ACCESS_KEY=xxxxxxxxxxxxxxxxxxxxxxxx
GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxxdocker compose build --no-cache
docker compose upExpected output — you should see this:
[+] Building 22.4s (8/8) FINISHED
✔ Image devmate-backend Built
✔ Container devmate_v2 Started
devmate_v2 | INFO: DevMate v2 starting up
devmate_v2 | INFO: DO_GRADIENT_API_KEY detected ✓
devmate_v2 | INFO: GITHUB_TOKEN detected (higher rate limits) ✓
devmate_v2 | INFO: Application startup complete.
devmate_v2 | INFO: Uvicorn running on http://0.0.0.0:8000
http://localhost:8000
- Paste this URL into the input box:
https://github.com/pallets/flask - Click Index Repo
- Wait ~30 seconds. You should see:
✓ Repository indexed successfully. 212 files loaded · Framework: Flask · Dependencies: 28 packages detected
[CHAT] tab — type:
How does Flask handle routing?
Expected: Detailed explanation with code examples using Flask's actual @app.route() decorator.
[BUILD] tab — type:
Add a health check endpoint to Flask
Expected: Code blocks with modify/create action badges, file paths validated against the real repo, copy buttons per block.
[DEBUG] tab — paste:
AttributeError: 'NoneType' object has no attribute 'json'
Expected: Root cause diagnosis, fix code referencing real Flask files, prevention tip.
DevMate v2 reads your entire GitHub repository and becomes an AI assistant that knows your specific codebase — not generic answers.
| Mode | What it does |
|---|---|
| [CHAT] | Ask anything about the codebase — architecture, patterns, "how does X work" |
| [BUILD] | Describe a feature → get exact code matching your folder structure, naming conventions, and pinned dependency versions |
| [DEBUG] | Paste error + stack trace → root cause diagnosis using your actual indexed code |
GitHub URL
↓
Fetch all source files concurrently (filtered, up to 100KB each)
↓
Parse dependencies — requirements.txt / package.json / go.mod / Cargo.toml
↓
Detect code patterns — framework, naming conventions, import style, test framework
↓
Build rich system prompt — file tree + deps + conventions + entry points
↓
DO Gradient Serverless Inference — llama3.3-70b-instruct
↓
Chat / Generate / Debug with full repo context
| Feature | How DevMate uses it |
|---|---|
| Serverless Inference | All AI — chat, code generation, debugging — via llama3.3-70b-instruct at https://inference.do-ai.run/v1/chat/completions |
| Model Access Keys | Secure key-based auth for the inference endpoint |
| Knowledge Base | Architecture designed for DO Gradient KB vector search; uses prompt injection for broad compatibility |
| Agent pattern | System prompts inject full repo context — file tree, pinned deps, naming conventions, entry points |
| Feature | DevMate v2 | Cursor | GitHub Copilot | ChatGPT |
|---|---|---|---|---|
| Full repo indexing | ✅ | Partial | ❌ | ❌ |
| Pinned dep version awareness | ✅ | ❌ | ❌ | ❌ |
| File path validation | ✅ | ❌ | ❌ | ❌ |
| Naming convention matching | ✅ | ❌ | ❌ | ❌ |
| Self-hosted on DO infrastructure | ✅ | ❌ | ❌ | ❌ |
| Debug with repo context | ✅ | Partial | ❌ | ❌ |
| No IDE install required | ✅ | ❌ | ❌ | ✅ |
| Python + Node + Go + Rust deps | ✅ | ❌ | ❌ | ❌ |
devmate/
├── backend/
│ ├── main.py # FastAPI app — 8 routes
│ ├── github_client.py # GitHub REST client + dep extraction
│ ├── gradient_agent.py # DO Gradient Inference — chat/generate/debug
│ ├── gradient_kb.py # Knowledge Base client (prompt injection mode)
│ ├── gradient_inference.py # Serverless Inference client
│ ├── code_generator.py # Prompt builder + file path validator
│ ├── docs_fetcher.py # 80+ versioned docs URL registry
│ └── session_store.py # In-memory session store (4h TTL)
├── frontend/
│ └── index.html # Dark terminal UI — single file, vanilla JS
├── Dockerfile
├── docker-compose.yml
├── .env.example
└── DEMO_SCRIPT.md
| Method | Route | Description |
|---|---|---|
POST |
/index |
Start indexing a GitHub repo |
GET |
/status/{id} |
Poll indexing progress |
POST |
/chat |
Ask a question |
POST |
/generate |
Generate code for a feature |
POST |
/debug |
Diagnose error + stack trace |
GET |
/health |
Health check |
Interactive docs: http://localhost:8000/docs
| Problem | Fix |
|---|---|
ERR_EMPTY_RESPONSE |
Run docker logs devmate_v2 --tail 20 and check for errors |
DO_GRADIENT_API_KEY not set |
Check .env exists in the devmate folder with no spaces around = |
| Port 8000 already in use | docker stop devmate_v2 && docker rm devmate_v2 then docker compose up |
| Indexing stuck at "Initializing" | Repo may be private — use a public repo like pallets/flask |
| Large repo times out | Add GITHUB_TOKEN to .env — raises rate limit from 60 to 5000 req/hr |
# Stop
docker compose down
# Start again (fast — uses cached build)
docker compose up
# Full rebuild after any code change
docker compose build --no-cache && docker compose upMIT
