GitHub discovery

Promising new AI projects

Young, active repositories worth an early look.

How this list is chosen

Candidates are the repositories GitHub's topic search returns for "llm" and "local-llm" when we collect, about three times a day. A repository must have been observed in the last 3 days to be listed, and each view lists at most 12.

Figures as of our latest count, . We have stored star counts for 137 repositories since .

  1. aosama/astronomical

    Set one RAM ceiling. Run bigger sparse language and vision models on Apple Silicon — experts stream from SSD when they don't fit, without hidden quantization. https://aosama.github.io/astronomical/ Join Discord Server https://discord.gg/dc4E6r4WD

    Licence
    Apache-2.0
    Last activity
    Last push
    Latest release
    v0.2.77,
    Stars
    22 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Shipped release v0.2.77 on 17 Oct 2026.
    • Created 3 Aug 2026, last push 18 Oct 2026.
    • Tagged apple-silicon, local-llm, macos, mlx.

  2. runapex/apex-router

    Adaptive model routing that learns the best model per task class behind a statistically sound promotion gate — plus a measure-first local-model offload subsystem (gated codegen, review pre-filter), grounded code-Q&A, and cache-cost tooling (cache-read reports, prefix budgets, handoff nudges). Pure-…

    Licence
    MIT
    Last activity
    Last push
    Latest release
    None published (checked )
    Stars
    1 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Created 1 Aug 2026, last push 18 Oct 2026.
    • Tagged anthropic-claude, code-search, codex, cost-optimization.

  3. jeremystevens/beyond

    BEYOND: Built Entirely on Your Own Network & Device. A local AI assistant with memory, tools and games: what Local AI Studio users wanted next.

    Licence
    MIT
    Last activity
    Last push
    Latest release
    None published (checked )
    Stars
    0 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Created 17 Oct 2026, last push 18 Oct 2026.
    • Tagged ai-assistant, chatbot, fastapi, llama-cpp.

  4. Naokuma005288/KokoroNe-AI

    AIが回答の中に、動かせる画面を作る。Windows 11向けAIアシスタント（α版・無料・ローカルAI対応）

    Licence
    Apache-2.0
    Last activity
    Last push
    Latest release
    None published (checked )
    Stars
    0 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Created 1 Oct 2026, last push 18 Oct 2026.
    • Tagged ai-assistant-app, ai-assistant-code, ai-assistant-free, ai-assistant-tools.

  5. sunilteja93/llmrig

    Know what your rig can run. Open compatibility and performance intelligence for local AI.

    Licence
    MIT
    Last activity
    Last push
    Latest release
    v0.9.3,
    Stars
    1 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Shipped release v0.9.3 on 16 Oct 2026.
    • Created 19 Aug 2026, last push 18 Oct 2026.
    • Tagged apple-silicon, benchmarking, hardware-compatibility, llama-cpp.

  6. kyleslight/shun

    The harness that makes consumer-GPU models feel first-class. Local-first: no telemetry, no account, no cloud tier.

    Licence
    MIT
    Last activity
    Last push
    Latest release
    v0.1.126,
    Stars
    7 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Shipped release v0.1.126 on 18 Oct 2026.
    • Created 21 Aug 2026, last push 18 Oct 2026.
    • Tagged ai-agent, coding-agent, desktop-app, developer-tools.

  7. Pigeon-Home/homing-pigeon

    Self-hosted, SI-controlled local agent. Tell your cloud SI (Grok, ChatGPT, etc.) to send signed jobs via email or a cloud queue, and they execute on your machine: Linux, macOS, FreeBSD, Windows or any container. An oracle on the host watches to ensure nothing bad or unintended occurs.

    Licence
    Apache-2.0
    Last activity
    Last push
    Latest release
    v3.7.22,
    Stars
    1 on
    Growth
    +0 stars over 7 hours, from to (0 a day)

    Star history

    Why look

    • Shipped release v3.7.22 on 17 Oct 2026.
    • Created 13 Oct 2026, last push 18 Oct 2026.
    • Tagged ai-agents, automation, email, intrusion-detection.

  8. clay-good/anvilate

    Anvilate is an open-source, local-first design agent for mechanical engineers. Describe a part in plain English and receive a physics-validated, parametric STEP or DXF file that drops straight into CATIA, SolidWorks, NX, or AutoCAD, alongside the editable Python source code.

    Licence
    MIT
    Last activity
    Last push
    Latest release
    None published (checked )
    Stars
    10 on
    Growth
    +0 stars over 7 hours, from to (0 a day)

    Star history

    Why look

    • Created 8 Jul 2026, last push 18 Oct 2026.
    • Tagged ai-design-agent, automated-validation, b-rep-geometry, build123d.

  9. tejas4patel/dragonfarm

    Raise your own dragon: fine-tune a language model on your own documents, from R. Writes the training data, trains on your laptop, RAG built in.

    Licence
    A licence GitHub could not identify — read it
    Last activity
    Last push
    Latest release
    v0.3.4,
    Stars
    13 on
    Growth
    +0 stars over 7 hours, from to (0 a day)

    Star history

    Why look

    • Shipped release v0.3.4 on 16 Oct 2026.
    • Created 14 Sept 2026, last push 18 Oct 2026.
    • Tagged fine-tuning, llm, local-llm, lora.

  10. praxiscad/praxiscad-skill

    AI skill + MCP setup for PraxisCAD: local-first text-to-CAD for any MCP client, local LLMs included. Parametric parts, STEP, 3D printing, assemblies, drawings, DFM checks. Offline and private.

    Licence
    A licence GitHub could not identify — read it
    Last activity
    Last push
    Latest release
    v0.15.0,
    Stars
    0 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Shipped release v0.15.0 on 17 Oct 2026.
    • Created 9 Oct 2026, last push 18 Oct 2026.
    • Tagged 3d-printing, agent-skills, ai-cad, cad.

  11. majorminorlabs/research-evidence-classifier

    E028 Qwen3-8B evidence-classifier code, metrics, and provenance release; model weights withheld pending license review.

    Licence
    A licence GitHub could not identify — read it
    Last activity
    Last push
    Latest release
    None published (checked )
    Stars
    0 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Created 24 Sept 2026, last push 18 Oct 2026.
    • Tagged evidence-classification, local-llm, major-minor, research.

  12. psewar/skirnir

    Skirnir - role- and state-aware router for Ollama: roles instead of model names, busy detection for gaming PCs, scheduler, client identities, cloud tiers, decision engine, Home Assistant integration

    Licence
    MIT
    Last activity
    Last push
    Latest release
    v0.13.0,
    Stars
    1 on
    Growth
    Not measured yet: we have counted its stars once, and have been measuring it since .

    Why look

    • Shipped release v0.13.0 on 5 Oct 2026.
    • Created 18 Sept 2026, last push 18 Oct 2026.
    • Tagged go, gpu, home-assistant, homelab.