Navni AI — AI Code Assistant

Free, private, local-first AI assistant for VS Code. Use Ollama for 100% private coding, or bring your own API key for cloud providers. No telemetry, no subscriptions.

Version 1.0.0 | Changelog | GitHub

Why Navni AI?

🔒 100% Private — Run locally with Ollama. Your code never leaves your machine.
💰 Free Forever — No subscription required. Bring Your Own Key for cloud APIs.
⚡ Inline Completions — Tab-autocomplete ghost text, like Copilot, but local and free.
🧠 Context Engine — Background indexing with semantic search across your entire project.
🤖 6 Specialized Agents — Agent, Chat, Generator, Reviewer, Tester, Debug Helper.

Features

⚡ Inline Code Completions (Tab-Autocomplete)

Ghost text suggestions as you type — press Tab to accept
Uses fast FIM (Fill-in-Middle) models for instant completions
Separate model setting — use a small 1.3B model for speed while chat uses a larger model
Debounced, cached, and cancellation-aware

Streaming responses with smooth "Thinking..." animation
Direct Ollama streaming — bypasses CLI for ultra-low latency (~200ms first token)
Markdown rendering with syntax-highlighted code blocks
Mermaid diagram rendering — AI can generate visual diagrams inline
Stream metadata — see model name, token count, and tok/s after each response
Stop generation button to cancel anytime

🔧 Smart Apply & Ripple Detection

Smart Apply — Apply code from chat to files using VS Code's native diff editor
Next Edit detection — When you apply a change, Navni finds all callers/usages and suggests bulk updates
File Operations — Create, modify, and delete files directly from chat

🌿 Conversation Branching

Edit previous messages to fork into new threads
Navigate branches with left/right arrows
Full history preserved — never lose context

🧠 Context Engine

Background indexing with file watchers for real-time updates
Semantic search across your entire project
Auto-attach active editor context on every message
Status bar indicator showing index health

🤖 Multi-Agent System

Agent	Purpose
⚡ Agent	General-purpose with file operations and workspace awareness
💬 Chat	Lightweight conversations without file context
⚡ Generator	Create new code, scaffolds, and boilerplate
🔬 Reviewer	Code analysis, security review, best practices
🧪 Tester	Generate unit tests and test plans
🐛 Debug Helper	Stack trace analysis, error fixing in 7 languages

📊 Additional Features

Agentic Task Planning — Complex requests are auto-decomposed into steps
Project Intelligence — Auto-detect frameworks, tech stack, architecture
Multi-Session Chat — Multiple independent sessions with persistence
Diff Engine — Preview all changes before applying
MCP Integration — Connect to Model Context Protocol servers

Supported Providers

Provider	Type	Cost
Ollama	Local	Free
Groq	Cloud	Free tier
Google Gemini	Cloud	Free tier
OpenAI	Cloud	Paid
OpenRouter	Cloud	Pay-per-use
Cerebras	Cloud	Free tier

Quick Start

1. Install the Navni CLI

pip install cognify-code

2. Set up a model

# Local (free, private)
ollama pull deepseek-coder:6.7b

# For inline completions (optional, recommended)
ollama pull deepseek-coder:1.3b

3. Install this extension and start coding!

The sidebar opens automatically. Type a message or just start coding to see inline completions.

Keyboard Shortcuts

Command	Windows/Linux	Mac
Review File	`Ctrl+Shift+R`	`Cmd+Shift+R`
Generate Code	`Ctrl+Shift+G`	`Cmd+Shift+G`
Explain Selection	`Ctrl+Shift+E`	`Cmd+Shift+E`
Open Chat	`Ctrl+Shift+C`	`Cmd+Shift+C`

Right-click selected code for Review Selection, Explain Selection, and Edit Code with AI.

Settings

Setting	Description	Default
`cognify.provider`	LLM provider	`ollama`
`cognify.model`	Chat model	`deepseek-coder:6.7b`
`cognify.completion.enabled`	Enable inline completions	`true`
`cognify.completion.model`	Completion model (fast, small)	(uses chat model)
`cognify.completion.debounceMs`	Typing pause before completing	`300`
`cognify.autoContext`	Auto-include relevant files	`true`
`cognify.maxContextTokens`	Max context tokens	`8000`
`cognify.cliPath`	Custom CLI path (Android/Termux)	(auto-detect)

Troubleshooting

"cognify not found"

pip install cognify-code
cognify --version

Slow completions

Use a small model for completions: set cognify.completion.model to deepseek-coder:1.3b
Lower cognify.completion.debounceMs to 200

Connection errors

cognify status

Contributing

Contributions welcome! Visit our GitHub repository.\n

License

MIT License — see LICENSE for details.

NavniAi