Introducing Agentantra: Sovereign AI in a 9MB Binary

Today, on India's 80th Independence Day (Swatantrata Diwas), we are proud to open-source Agentantra (v0.1.0) — India's first compiled, zero-dependency sovereign AI agent engine.

For years, building autonomous AI agents required wrestling with heavy Python virtual environments, fragile dependency trees, multi-gigabyte Docker images, and slow interpreter boot times.

Agentantra replaces this entire operational nightmare with a single 9MB compiled Go executable.

Key Benchmarks & Engineering Highlights

• 9MB Single Binary: Download one static binary, put it in your PATH, and start executing agents immediately.
• <10ms Cold Starts: Instant initialization allows agents to spin up per-request without persistent container idle waste.
• 15MB Base Memory Footprint: Run hundreds of sovereign agents concurrently on a standard cloud VM or edge machine.
• Native MCP Host: Built-in Model Context Protocol client multiplexing over standard stdin/stdout with zero CGO dependencies.
• 100% Data Residency: Session memory, vector indexes, and execution logs reside locally or within Indian sovereign clouds.

30-Second Quickstart

Install and run your first sovereign agent with a single shell pipeline:

The Declarative liate.json Standard

Every agent in Agentantra is defined by the 5-pillar EkaMantra specification:

{
  "L": "sarvam/sarvam-105b",
  "I": {
    "memory": "sessions/time-agent-india-v4"
  },
  "A": {
    "name": "time-agent",
    "intent": "You are a time analyst. When asked about time in India or IST, pass timezone='Asia/Kolkata'.",
    "skills": "./skills.md"
  },
  "T": {
    "mcp-server-time": {
      "command": "uvx",
      "args": ["mcp-server-time"]
    }
  },
  "E": {
    "SARVAM_API_KEY": "$SARVAM_API_KEY",
    "MAX_TURNS": "10"
  }
}