Akash deploy bot

**Akash Deploy Agent: one sentence to a live URL on decentralized cloud.**

Comment

Inspiration

Deploying to Akash requires understanding SDL files, blockchain escrow, provider bidding, and lease management. Vercel made centralized deployment feel like magic. I wanted the same thing for decentralized cloud.

What It Does

Type "deploy nginx on port 80" → get a live URL on Akash. That's it.

The agent handles SDL generation, marketplace submission, bid selection, and lease creation. It also lists and tears down deployments. The entire stack runs on Akash — the agent, the LLM (Llama 3.3 70B on AkashML), and every deployment it creates.

It deployed itself to Akash using its own pipeline.

How I Built It

  1. Core library (akash_core.py) — 350 lines of Python wrapping the Akash Console Managed Wallet API. Three functions: deploy(), list_deployments(), close_deployment(). Template-based SDL generation (100% valid every time, unlike LLM-generated IaC which fails ~70%).

  2. AI agent (agent.py) — FastAPI app with chat UI. User messages go to Llama 3.3 70B on AkashML via OpenAI-compatible function calling. The LLM selects which tool to invoke; Python formats the results directly (no wasted second LLM call).

  3. Self-hosted — Containerized, published to Docker Hub via GitHub Actions, deployed to Akash.

Key distinction from Akash's own MCP server: Their MCP server exposes low-level blockchain primitives requiring a crypto wallet and multi-step orchestration across 12+ separate tools. My library abstracts the entire deploy lifecycle into a single function call with a credit card API key — no crypto knowledge needed.

The core library is framework-independent. It works as a CLI tool, a Claude Code skill, an OpenClaw skill, or a tool in any agent framework. The web agent is one wrapper; the deployment capability is portable.

Challenges

  • Akash API latency (12-26s per call) — mitigated with progress indicators and narration
  • AkashML tool calling quirks — empty argument strings, overeager tool invocation on general questions
  • Performance — eliminated unnecessary LLM round-trips, cutting response time from 22s to ~6s for non-deployment commands

What's Next

  • Docker Compose support (deploy multi-service apps)
  • Build from source (Dockerfile → image → Akash, fully automated)
  • Cost monitoring and escrow alerts
  • Published integrations for Claude Code, OpenClaw, and MCP

Built With

Share this project:

Updates