Skarta
Build agents in Python. Ship them on a Rust runtime.
Start new thread
Chirotpal

Skarta - Build agents in Python. Ship them on a Rust runtime.

by•
Skarta is a production-grade Rust runtime for multi-agent systems. Write agents in Python, ship them on a Rust binary that owns scheduling, validation, sessions, budgets, telemetry, sandboxing, and human-in-the-loop approvals. One install. Apache 2.0.

Add a comment

Replies

Best
Chirotpal
Maker
📌
Hi Product Hunt, Chirotpal here, the maker. Skarta is a Rust runtime for multi-agent AI systems that you write in Python. The short story: every agent framework I tried was a Python library. Beautiful for a demo. Brittle the day a real customer used it. A single crash took the whole workflow down. Retries forgot their place. Spend tracking was whatever print statements I happened to add. Observability was a weekend project on top. Skarta is the opposite shape. One pip install skarta ships a Rust binary that runs under your Python code. The runtime owns the boring, hard parts: - Auto-DAG from your data shapes, no flow charts to wire. - Hard budget caps the runtime enforces, a runaway prompt cannot blow up your OpenAI bill. - Session memory that survives restarts, with branching and named checkpoints (try a tangent, snap back if you do not like it). - In-runtime HTTP for webhooks, with duplicate POSTs dropped for you, no Flask or FastAPI in front. - Human-in-the-loop approval gates on any tool, Slack or email out of the box. - Per-tool sandboxing for filesystem, network, and env vars, runtime-enforced. - OpenTelemetry traces and Prometheus metrics in the same wheel. - And more Works with Anthropic, OpenAI, OpenRouter, and any OpenAI-compatible endpoint (Ollama, vLLM, Azure OpenAI, Groq). Python SDK ships today. Apache License 2.0. No hosted tier, no signup wall, no "contact sales" CTA hiding the good bits. Try it in 60 seconds: pip install skarta export OPENAI_API_KEY=sk-... Then a two-agent typed pipeline that actually calls a model is 20 lines of Python (full snippet in the README). A few things I would love the PH community's eyes on: - The "From your first agent to a production service" 11-step ladder in the README. Which rung is the one you would have actually used at your job? - Anywhere the docs read like docs-for-the-author, not docs-for-you, please call it out. - What is the next integration that would tip Skarta from "interesting" into "I am using this on Monday"? I will be here all day. Roast it, break it, ask anything. Repo: https://github.com/SarthiAI/Skarta