Categories
Self-Hosted AI Agents MCP Server Local LLM Personal Assistant Coding Agent Voice Assistant AI CompanionStars
Forks
Watchers
Developer links
Sara
Meet Sara, an AI companion with a personality you design and a coding partner that writes, runs, and tests real software, using any model you choose. Pick a name and pronouns, start from a character or set dials for playfulness and bluntness, and keep memories you can inspect, correct, and forget. Chat from the web app, Windows and Mac apps, Android, iPhone, Telegram, Discord, Signal, or email, and talk through live voice calls you can interrupt or video calls with her moving portrait. A planner turns plain sentences into to-dos, reminders, and notes, with a board, a Today home screen, and routines such as a morning briefing. The library reads PDFs, scans, Word, Excel, PowerPoint, and e-books, then answers with the document and page it came from. For code, each project gets a sandboxed workspace with files, a shell, and git; Sara asks before risky commands, shows test results as cards, and lists every command and its output under each reply. Goals finish only when your own check passes. Use Ollama, vLLM, or llama.cpp locally, connect OpenAI, Anthropic, xAI, or OpenRouter, or reuse the ChatGPT and Claude plans you already pay for. Connect coding agents over MCP, call an OpenAI-compatible API, and add plugins from a signed store. Business workspaces add passkeys, roles, and shared workspaces. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.
Benefits
- A companion with memory you control
- Memories of what you told her are visible, correctable, and forgettable, and a personality built from simple dials or a starter character stays consistent across web, phone, Telegram, and Discord chats.
- Coding that shows what ran
- Each reply carries a line listing every command and its output, a check compares what she says she did with what actually ran, and goals end only when your own check passes.
- Any model, local or cloud
- Connect Ollama, vLLM, or llama.cpp locally, OpenAI, Anthropic, xAI, or OpenRouter in the cloud, or reuse a ChatGPT, Claude Pro or Max, or SuperGrok plan you already pay for.
- Small enough for a modest VPS
- With a cloud model or a subscription, Sara needs only 1 GB of free RAM and one CPU core on x86_64 or arm64, and ships as a Docker image for both.
Features
- Sandboxed coding projects
- Each project gets a workspace with files, a shell, and git; Sara asks before running commands, shows test results as cards, and can try risky commands in a sandbox copy.
- Personality and editable memory
- Dials or starter characters set her personality, moods and running jokes add character, and a memory map shows what she remembers so you can correct or forget it.
- Planner, Today, and routines
- To-dos, notes, and reminders come from plain sentences, with a board, a Today home screen, and a routines gallery covering morning briefings and weekly reviews.
- Document library with sources
- Reads PDFs including scans and books of 500 to 1,000 pages, plus Word, Excel, PowerPoint, and e-books, and each answer names the document and page.
- Voice, video, and chat bridges
- Live voice calls you can interrupt, video calls with her moving portrait, Hey Sara hands-free on Windows, plus Telegram, Discord, Signal, and email bridges.
- MCP, API, and add-ons
- An OpenAI-compatible API and MCP server connect Claude Code, Cursor, and SillyTavern, while plugins, skills, and tool servers install from the signed Sara Store behind review cards.