Yggopen-jarvis/OpenJarvis
↗ GitHubPersonal AI, On Personal Devices
2,041
Stars
362
Forks
24
Watchers
15
Open Issues
Safety Rating A
No hardcoded secrets, malicious code patterns, suspicious dependencies, or prompt injection attempts were detected. The repository is a legitimate academic/open-source framework from Stanford research labs with clear provenance, an Apache 2.0 license, and transparent documentation. The installation flow relies on standard Python tooling (uv, maturin) and well-known open-source inference backends.
ℹAI-assisted review, not a professional security audit.
AI Analysis
OpenJarvis is a local-first personal AI agent framework developed at Stanford's Hazy Research and Scaling Intelligence Lab. It provides a software stack for building on-device AI agents that run locally by default, calling cloud APIs only when necessary. The framework ships with shared primitives for building on-device agents, an evaluation system that treats energy, FLOPs, latency, and cost as first-class constraints alongside accuracy, and a learning loop for improving models using local trace data. It supports multiple local inference backends (Ollama, vLLM, SGLang, llama.cpp) as well as cloud providers, and includes a Rust extension for performance-critical components.
Use Cases
- Building and running personal AI agents that execute locally on consumer hardware
- Evaluating local LLM performance with energy, latency, cost, and accuracy metrics
- Developing research platforms for on-device AI efficiency (Intelligence Per Watt)
- Deploying a local-first AI assistant via CLI or FastAPI server
- Fine-tuning and improving models using locally collected inference trace data
Tags
Project Connections
local-talking-llm
→local-talking-llm provides a fully local voice interface layer (STT + TTS) that could be integrated with OpenJarvis agents to add voice interaction capabilities to the local-first AI stack.
skyclaw
→Both are local-first AI agent frameworks with multi-modal capabilities, persistent runtimes, and support for multiple LLM backends, targeting similar use cases of autonomous personal AI on local hardware.
clawvault
→ClawVault's local-first markdown-native memory and knowledge graph system could serve as a persistent memory backend for OpenJarvis agents, complementing its agent primitives with structured long-term context.
kavach
→Kavach's EDR capabilities for monitoring and restraining local AI agents could be used alongside OpenJarvis deployments to provide security oversight and audit logging for on-device agent operations.
code-review-graph
→code-review-graph's MCP-based context engineering and blast-radius analysis tooling could serve as a specialized tool within OpenJarvis agent workflows for software development tasks.