No comments yet.
Watch 45s demo: https://asciinema.org/a/p5mzDGAoHTUHNEaVeROHpFibx
---
QUICK START
# Install
pip install solveig
# Run from local models or remote APIs
solveig -u "http://localhost:5001/v1" "Create a demo BlackSheep webapp"
# Mix config files and CLI args
solveig -c solveig.config -k "<API_KEY>" -m "gpt-5"
---
FEATURES
AI Terminal Assistant - Automate task planning, file management, code analysis and system management using natural language in your terminal.
Safe by Design - Granular controls with pattern-based permissions. File operations prioritized, and shell commands can be disabled.
Plugin Architecture - Extend capabilities through drop-in plugins. Add SQL queries, web scraping or block dangerous commands with 100 lines of Python.
Modern CLI - Clear interface with task planning and listing, file content previews, diff editing, API usage tracking, code linting, waiting animations and rich tree displays for informed user decisions.
Provider Independence - Works with any OpenAI-compatible API, including local models.
tl;dr: similar idea to Claude Code (https://claude.com/product/claude-code) or Aider (https://aider.chat/), focusing on providing explicit user consent, granular configuration, drop-in plugins and the ability to integrate any model, backend or API.
See the Features for more: https://github.com/FSilveiraa/solveig/blob/main/docs/about.m...
---
TYPICAL TASKS
- "Find and list all the duplicate files inside ~/Documents/" - "Check my essay Final.docx for spelling, syntax or factual errors while maintaining the tone" - "Refactor my test_database.ts suite to be more concise" - "Try and find out why my computer is slow" - "Create a dockerized BlackSheep webapp with a test suite, then build the image and run it locally"
---
So it's a coding assistant?
You can use Solveig for analyzing, editing and testing your code, and all of these scenarios have received significant support through development features like code linting. But I didn't build Solveig with a single kind of use case in mind.
---
So it's yet another LLM-in-my-terminal?
Sort of. Solveig tries to do a few things that other tools don't, and to do the shared features with clearer UX, explicit consent, and deeper configuration. It's not an IDE extension, doesn't require a GUI, and it's not built for a specific user type or scenario.
At the same time, Solveig's competitors are mature projects with real user testing that you should check out. I've written a detailed comparison (https://github.com/FSilveiraa/solveig/blob/main/docs/compari...) to similar tools in the market in the docs.
---
UPCOMING
I have a Roadmap (https://github.com/FSilveiraa/solveig/discussions/2) available, and feel free to suggest new features or improvements. I've recently added user-defined system prompt templates, and now I'm working on adding token counting from API messages instead of relying on encoders.
---
A cool aspect of this project is that I can use Solveig to analyze and improve its own code itself, which also gives me a lot of exposure to its actual usability.
I appreciate any feedback or comment, especially anyone who can try out Solveig using Anthropic or Gemini APIs. Tell me if it helped you do something or what stopped you from using it properly in your specific case. Even if you can't see how Solveig could help you let me know, that's an issue with me communicating value that I need to fix.
Leaving a star on the repository (https://github.com/FSilveiraa/solveig) is also very much appreciated.