← OSS.Radar home

Powered by aegismemory.com · Aegis Memory repository

Underrated AI open-source gems

Lower-visibility AI repositories with strong quality and maintenance signals.

guqiong96/Lvllm

LvLLM is a special NUMA extension of vllm that makes full use of CPU and memory resources, reduces GPU memory requirements, and features an efficient GPU parallel and NUMA parallel architecture, suppo

Category
model serving
Stars
428
Readiness
ready (95/100 heuristic points; not a probability)
Data confidence
high
Maintainer health
healthy
Maintenance risk
0/100 · high confidence

Why: +38 stars in 7 days; 100+ commits in 30 days

Why it may be a gem: consistent human and community activity; healthy maintenance and project fundamentals

Strongest signals: push recency, commit activity, contributor breadth. Risks: None identified. Missing inputs: None.

Capped lower bounds: 30-day commits, lifetime contributors.

cpu decode gpu hybrid inference model

cosmo-wander-ai/cosmo-edge

Production-grade C++ edge AI engine for video analytics with visual pipeline orchestration, on-device VLM, and real-time OSD.

Category
local and edge ai
Stars
424
Readiness
ready (93/100 heuristic points; not a probability)
Data confidence
high
Maintainer health
healthy
Maintenance risk
8/100 · high confidence

Why: +84 stars in 7 days; 100+ commits in 30 days

Why it may be a gem: healthy maintenance and project fundamentals; consistent human and community activity

Strongest signals: push recency, commit activity, issue load. Risks: no pull-request review responses in 30 days. Missing inputs: None.

Capped lower bounds: 30-day commits.

bm1688 computer-vision cpp17 edge-ai grounding-dino on-device-ai

Ikalus1988/MisakaNet

📚 A zero-dependency, git-backed micro-lesson library for AI Agents to asynchronously share and search verified debugging experience. Python stdlib only. | https://misakanet.org

Category
agent framework
Stars
419
Readiness
ready (97/100 heuristic points; not a probability)
Data confidence
high
Maintainer health
healthy
Maintenance risk
0/100 · high confidence

Why: +78 stars in 7 days; 100+ commits in 30 days

Why it may be a gem: healthy maintenance and project fundamentals; consistent human and community activity

Strongest signals: push recency, commit activity, contributor breadth. Risks: None identified. Missing inputs: None.

Capped lower bounds: 30-day commits, response activity.

agent-framework agent-network ai-agent claude devops distributed-memory

MunnaXbadmash/ai-dev-assistant-framework

A plug-and-play framework for AI-assisted software development, enhancing context-aware collaboration in complex codebases. Perfect for tools like GitHub Copilot. 🐙✨

Category
agent framework
Stars
5
Readiness
ready (87/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal agent framework project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai ai-agent ai-agents-framework ai-framework ai-governance claude-ai

anwen0724/megumi

Local-first desktop coding agent for Windows with BYOK models, visible tool execution, approval controls, and session branching.

Category
ai coding and assistants
Stars
5
Readiness
ready (87/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai coding and assistants project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-agent ai-coding byok code-assistant coding-agent desktop-app

roberto729a/OllamaRAG

🤖 Build a smart AI assistant that learns from any website using a Retrieval-Augmented Generation framework with local models powered by Ollama.

Category
rag and search
Stars
6
Readiness
ready (96/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal rag and search project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-framework beautifulsoup chromadb gemma-2b knowledge-base langchain

IamNotShady/LLMIngress

Self-hosted AI gateway for agents — route providers, virtual models, limits, and usage from one console.

Category
gateways and routing
Stars
7
Readiness
ready (87/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +1 stars in 7 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai ai-agents ai-gateway anthropic claude claude-code

Providex-AI/rootsign

Rootsign is an open-source tamper-evident decision and action provenance logging library for AI agents

Category
agent framework
Stars
8
Readiness
ready (87/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +2 stars in 30 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agent-accountability agents ai-agents audit-trail compliance crewai

Mizistein/omlx

🤖 Optimize LLM inference on Mac with continuous batching and SSD caching managed from your menu bar for efficient performance.

Category
model serving
Stars
8
Readiness
ready (87/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal model serving project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

apple-silicon chatbot inference-server llm macos mlx

Indexical-Metrics-Measure-Advisory/watchmen

Watchmen Platform is a low code data platform for data pipeline, meta data management , analysis, indicator objective analysis and quality management

Category
vector dbs and data
Stars
8
Readiness
ready (100/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal vector dbs and data project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, fork interest. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

charts data-pipeline data-quality data-quality-monitoring data-visualization indicator

psyb0t/aigate

A self-hosted AI platform — inference, tool use, browser automation, image generation, speech synthesis, transcription, object storage, agentic code execution, and more — behind a single OpenAI-compat

Category
gateways and routing
Stars
11
Readiness
ready (83/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +1 stars in 7 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai ai-gateway browser-automation claude-code docker docker-compose

transitive-champaign778/ApkClaw

Automate Android devices with an LLM agent that takes natural language commands from messaging apps and runs device tasks autonomously

Category
gateways and routing
Stars
13
Readiness
ready (95/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal gateways and routing project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

9patch agents android-security anthropic antivirus appshielding

EnragedAntelope/comfyui-identity-forge

A character forge for ComfyUI: dial in or randomize every detail - face, hair, body, wardrobe, mood - and get a coherent, reproducible person, canonically-described cosplayer, or creature as ready-to-

Category
multimodal media
Stars
15
Readiness
ready (97/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +2 stars in 7 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, fork interest. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

character-creator character-generator comfyui comfyui-custom-nodes prompt-generator text-to-image

archubbuck/workspace-architect

Workspace Architect is a zero-friction CLI tool that provides curated collections of specialized agents, instructions, and prompts to supercharge your GitHub Copilot experience.

Category
ai coding and assistants
Stars
17
Readiness
ready (96/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai coding and assistants project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, fork interest. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agentic-workflow ai-agents ai-assistant ai-personas automation chatmodes

krishnakaanthreddyy1510-cell/RedSOC

An adversarial evaluation framework for LLM-integrated Security Operations Centers

Category
ai security and guardrails
Stars
6
Readiness
ready (86/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai security and guardrails project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

adversarial-ml ai-security cybersecurity large-language-models llm-security prompt-injection

rayl15/Godwit

Run GPT-OSS-120B on a 16 GB Mac. Streams mixture-of-experts weights from SSD so only the 2 GiB trunk stays resident. Swift + Metal, no dependencies.

Category
local and edge ai
Stars
19
Readiness
ready (92/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal local and edge ai project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

apple-silicon gpt-oss inference llm local-llm macos

swobuforge/swobu

Local compatibility boundary for Claude Code, Codex and AI clients. Route across providers without silently losing tools, reasoning or session state.

Category
gateways and routing
Stars
6
Readiness
ready (78/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal gateways and routing project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-gateway aiproxy antrophic-compatible aws-bedrock azure-openai claude

Elliott-Crosby/Nodea

Branching AI chat canvas: fork any reply, compare branches, never lose context. ChatGPT alternative built on Claude. Free during beta.

Category
ai webui and interfaces
Stars
5
Readiness
ready (86/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai webui and interfaces project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-canvas ai-chat ai-tools anthropic branching-chat chatgpt-alternative

arsalanafzal010/SmartRAG

📄 Enable smart conversations with documents, images, and audio files using this advanced Retrieval-Augmented Generation system.

Category
multimodal media
Stars
5
Readiness
ready (86/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal multimodal media project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

autogen automation cli document-processing dotnet gpt

temiroff/Blacknode

A visual runtime for AI, GPU computing, and robotics, workflow framework for AI agents. MCP-native typed graphs, visual editor, Python export.

Category
agent framework
Stars
21
Readiness
ready (94/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +11 stars in 7 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agent-workflows ai-agents ai-tools anthropic crewai langgraph

defai-digital/ax-engine

One Mac process. Many models. Real speed. Multi-model LLM serving with prefix reuse, MTP acceleration, and OpenAI APIs — built for Apple Silicon, measured against mlx-lm and llama.cpp.

Category
local and edge ai
Stars
21
Readiness
ready (92/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal local and edge ai project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

apple-silicon embeddings gemma homebrew llm local-ai

asya-ai/asya-chat-ui

Better, open-source, LLM wrapper (user management, text processing, file processing, image generation, web search, local models)

Category
ai webui and interfaces
Stars
9
Readiness
ready (84/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai webui and interfaces project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai llm llm-ui llm-webui llms mcp

0xArchiveIO/sdk-python

Python client for 0xArchive: granular market data for Hyperliquid (perps, HIP-3, HIP-4, Spot) and Lighter.xyz.

Category
vector dbs and data
Stars
7
Readiness
ready (86/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal vector dbs and data project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

data-pipeline hyperliquid lighterxyz market-data python rest-api

eulogik/fugusashi

Fugusashi — Federated Learning of LLM Routing with Human-Interpretable Decisions. Open-source alternative to Sakana AI Fugu.

Category
gateways and routing
Stars
11
Readiness
ready (96/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +2 stars in 7 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, fork interest. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-orchestrator cma-es cost-optimization federated-learning llm llm-router

paperfoot/seedance-cli

ByteDance Seedance 2.0 / 2.5 from your terminal. Text-to-video, image-to-video, audio & video refs, up to 4K. One agent-ready Rust binary.

Category
multimodal media
Stars
23
Readiness
ready (89/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal multimodal media project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agent-cli agent-friendly ai ai-video bytedance byteplus

hermes-labs-ai/little-canary

Detects prompt injection by its effect on a sacrificial canary model, not just pattern matching: untrusted input hits a powerless model first, a behavioral check reads the residue, and it returns bloc

Category
ai security and guardrails
Stars
26
Readiness
ready (95/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai security and guardrails project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agent-security ai-agents ai-safety canary defense-in-depth detection

paraxaQQ/canary

Deterministic, read-only static auditor for GGUF models: behavioral chat-template backdoors + SSTI, tokenizer, config, and model-card surfaces - never renders, never loads weights. 192,032 repos, 137,

Category
ai security and guardrails
Stars
26
Readiness
ready (93/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal ai security and guardrails project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-security backdoor-detection chat-template gguf huggingface jinja2

arthi-arumugam-git/whatbroke

Diff your AI agent's behavior between two runs. See exactly which tool calls, args, costs and outputs changed when you swap models or edit prompts.

Category
evals and testing
Stars
15
Readiness
ready (90/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal evals and testing project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agent-testing agents ai ai-agents ai-tools anthropic

EasyMetaAu/helm-api

Self-hosted LLM gateway — route OpenAI/Anthropic/Gemini traffic by config, not code. Fallback, protocol translation, telemetry.

Category
gateways and routing
Stars
28
Readiness
ready (98/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: +1 stars in 7 days

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, fork interest. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

ai-gateway anthropic api-gateway claude gemini gpt

ROCK-LAB-PRIVATE-LIMITED/federaide

P2P multi agent orchestration harness, universal automation system and IDE

Category
multimodal media
Stars
6
Readiness
ready (86/100 heuristic points; not a probability)
Data confidence
low
Maintainer health
healthy
Maintenance risk
0/100 · low confidence

Why: High-signal multimodal media project

Why it may be a gem: strong signals despite limited visibility; healthy maintenance and project fundamentals

Strongest signals: push recency, issue load, documentation. Risks: None identified. Missing inputs: commit activity, contributor breadth, release recency, response activity, maintenance distribution.

agent-swarms agentic-ai cpp crossplatform deep-research-agent go