Krishna — Autonomous AI Developer Agent

Krishna is a modular, local-first AI agent with a chat interface, tools, RAG knowledge base, and step-by-step task execution.

Built with Next.js 15, LangChain, and OpenAI / Ollama.

For Teammates — Get Started in 3 Steps

Option A: Install via npm (recommended)

# 1. Install Krishna globally
npm install -g krishna-agent

# 2. Create your project
krishna init my-krishna

# 3. Configure LLM (pick one below) then start
cd my-krishna
krishna start

Option B: Clone from GitHub

git clone https://github.com/YOUR_USERNAME/krishna-agent.git
cd krishna-agent
npm install
cp .env.example .env.local
# Configure .env.local then:
npm run dev

Configure Your LLM

Open .env.local (copy from .env.example) and pick one:

OpenAI (cloud)

OPENAI_API_KEY=sk-your-key-here
OPENAI_MODEL=gpt-4o-mini

Ollama (local, free, no API key)

brew install ollama
ollama pull llama3.2          # ~2GB chat model
ollama pull nomic-embed-text  # for semantic search
ollama serve

USE_OLLAMA=true
OLLAMA_MODEL=llama3.2

Load Your Docs / Codebase

Krishna can answer questions about your code if you give it to her first:

# Index a repo
krishna ingest ./my-project

# Index docs or specific files
krishna ingest ./docs README.md

# Index a URL
krishna ingest https://docs.yourproject.com

# Check what's loaded
krishna ingest --stats

Krishna will automatically search the knowledge base when answering questions about your project.

CLI Commands

krishna init [dir]           # Create a new project
krishna chat                 # Chat with Krishna in the terminal ← new
krishna chat --simple        # Quick Q&A mode (no agent pipeline)
krishna start                # Start the web UI → http://localhost:3000
krishna build                # Production build
krishna ingest <path>        # Add docs/code to knowledge base
krishna status               # Show system + KB health
krishna help                 # All commands

Terminal Chat

$ krishna chat

  ██╗  ██╗██████╗ ██╗███████╗██╗  ██╗███╗   ██╗ █████╗
  ...

Session: a3f7c1d2  •  Mode: Agent  •  Server: http://localhost:3000

You [agent]  Give me a task or ask anything...

You [agent]  Calculate the 10 most common words in README.md

Krishna  🗂️ Creating execution plan...

Plan:
  1. Read the README.md file
  2. Execute code to count word frequency
  3. Return the top 10 words

▶ Step 1: Read README.md
  → tool: read_file
  ✓ done
▶ Step 2: Count words with code
  → tool: execute_code
  ✓ done
...

In-chat slash commands:

/agent      Switch to Agent Mode (full planning + tools)
/chat       Switch to Chat Mode (quick Q&A)
/ingest     /ingest ./src  — load files live
/clear      Clear terminal
/status     System health
/quit       Exit  (also Ctrl+C)

Features

| Feature | Description | |---------|-------------| | Agent Mode | Full Planner → Executor → Evaluator pipeline | | Chat Mode | Quick Q&A with streaming | | Knowledge Base | RAG from your docs/codebase | | 9 Tools | File I/O, web fetch, code execution, search | | Ollama | Run 100% locally, no API key | | Memory | Per-session + persistent long-term | | Streaming | Real-time step-by-step execution trace |

Architecture

User → Planner → Executor → Tools → Evaluator → Response
                     ↕
              Knowledge Base (RAG)
                     ↕
              Memory (short + long term)

See docs/architecture.md for the full breakdown.

Documentation

| Guide | Description | |-------|-------------| | Setup | Full install guide | | Architecture | How Krishna works | | Create Agent | Add tools and custom agents | | Prompt Engineering | Customize Krishna's behavior | | Testing | Test cases and benchmarks |

Publishing Your Own Version

# Update name/version in package.json
# Then publish to npm:
npm login
npm publish

# Teammates install with:
npm install -g your-package-name

License

MIT