Indusagi

Indusagi is an open-source developer framework built around a TypeScript SDK and a terminal-first Coding Agent CLI. It is released under the MIT license and available on both npm and PyPI, with no licensing fees for personal, institutional, or commercial use. The project is maintained by Varun Israni and hosted publicly on GitHub.

What It Is

Indusagi sits in the AI coding agent and agent framework category. Its core job is to give developers granular, code-level control over autonomous AI agent loops — including memory directories, tool execution, session compaction, and console UI rendering — rather than wrapping those concerns in a black-box platform. It ships two primary artifacts: the indusagi npm package (TypeScript SDK + embeddable library) and the indusagi-coding-agent CLI package (standalone terminal coding agent). Python equivalents (indusagi and induscode) are also available on PyPI.

Architecture and Capability Layers

The framework organizes its surface into named capability layers that can be imported individually or as a combined namespace:

• runtime — owns createAgent, the core agent loop entry point
• capabilities — assembles runnable tool sets via toolBox with presets: "read-only", "coding", or "all"
• llmgateway — model catalog, cost estimation, and provider routing
• TUI widgets — modular terminal dashboard components for custom console UI layouts

The CLI supports three operating modes: interactive REPL (default), one-shot print mode (-p), and a JSON wire/RPC protocol over stdio (--json / --rpc). Extensions are loaded at runtime via the jiti TypeScript compilation loader, so no manual pre-compilation step is needed.

Model and Provider Support

Indusagi integrates with cloud LLM providers — Anthropic, OpenAI, and Groq — and also supports fully offline, local inference through Ollama and Llama.cpp. The framework includes built-in client capabilities for the Model Context Protocol (MCP), allowing agents to connect to third-party tools, local databases, or external microservices that implement the protocol. Multiple MCP servers can be attached at CLI invocation time using the --mcp flag.

Session and Context Management

A key differentiator the project documentation highlights is automated session compaction: the CLI periodically compiles extended conversation histories and active branch logs into semantic text summaries. This preserves long-term project context across sessions while reducing API token consumption. Developers can also write custom hooks to intercept file operations and configure granular token boundaries.

Setup Path

The documentation states that initial environment setup takes under two minutes. The package requires Node.js 20 or newer and is ESM-only. Installing via npm install indusagi also provisions the indusagi CLI binary. A setup wizard connects to the preferred model provider, after which the agent is immediately usable. The Python packages follow the same pattern via pip install indusagi and pip install induscode.

Current Status

According to npm registry statistics cited on the homepage, the indusagi SDK package has accumulated approximately 8,700 total downloads in the past year (roughly 1,400 monthly), and the indusagi-coding-agent CLI package has approximately 10,500 total downloads in the past year (roughly 1,600 monthly). The project is actively maintained on GitHub and the documentation site is live at indusagi.com.