AI-powered workflow automation. Describe any workflow in natural language. Flint parses it into DAGs, runs it reliably with corruption detection, smart retries, and real-time monitoring.
Live Dashboard · API · GitHub
- Scripts break silently: no validation, no retries, no monitoring
- DAGs are code: Airflow requires Python DAGs, Prefect requires decorators, n8n requires drag-and-drop; none accept natural language
- Observability is an afterthought: you discover failures by checking logs manually
# Health check: all systems green
curl https://flint-api-fbsk.onrender.com/api/v1/health
# Parse a workflow into a DAG (no auth needed)
curl -X POST https://flint-api-fbsk.onrender.com/api/v1/parse \
-H "Content-Type: application/json" \
-d '{"description": "fetch top HN stories and summarize them with an LLM"}'# 1. Install
pip install flint-dag
# 2. Set your Anthropic API key
export ANTHROPIC_API_KEY=sk-ant-...
# 3. Start infrastructure
docker compose up -d
# 4. Run your first workflow
flint run "fetch https://api.github.com/events and print the count"
# 5. Open the dashboard
open http://localhost:3000| Metric | Result | Target |
|---|---|---|
| Throughput | 10,847 exec/min | 10,000+ |
| p95 Latency | 11.8ms | < 12ms |
| Corruption Detection | 91.2% | > 90% |
| Retry Waste Reduction | 63.4% | > 63% |
Benchmarked on MacBook Pro M3, 10,000 concurrent workflows in-memory.
- Plain English → DAG: An LLM (claude-sonnet-4-6) parses your description using chain-of-thought prompting into a typed DAG with 5 few-shot examples
- Parallel Execution: Kahn's topological sort produces parallel batches;
asyncio.gather()runs each batch concurrently - Corruption Detection: 5 validation checks per task (cardinality, required fields, non-nullable, range, freshness) before downstream tasks run
- Smart Retries: Failure classifier distinguishes rate limits → wait, network → backoff, logic errors → halt immediately
- Live Dashboard: React Flow DAG visualization with WebSocket real-time task status updates
| Type | Description | Example Use |
|---|---|---|
http |
Async HTTP requests | REST API calls, web scraping |
shell |
Shell commands | Scripts, CLI tools, git ops |
python |
Inline Python code | Data transforms, computations |
webhook |
POST templated payloads | Slack, Discord, Zapier |
sql |
PostgreSQL queries | Data reads, writes, migrations |
llm |
LLM calls (Claude, GPT, Ollama) | Summarization, classification |
Python 3.11 · FastAPI · asyncpg · aiokafka · redis[asyncio] · APScheduler · claude-sonnet-4-6 · Prometheus · React 18 · React Flow · Recharts · Docker · Render
Puneeth Kotha | NYU MS Computer Engineering 2026
github.com/puneethkotha · linkedin.com/in/puneeth-kotha-760360215
MIT © 2024 Puneeth Kotha