{"id":35066239,"url":"https://github.com/mr5niper/aegis-synthesis-architecture","last_synced_at":"2026-07-05T21:00:29.239Z","repository":{"id":325232416,"uuid":"1100234468","full_name":"Mr5niper/Aegis-Synthesis-Architecture","owner":"Mr5niper","description":"Sovereign, local-first personal AI assistant. Local LLM inference (llama.cpp) with multi-model hot-swap, RAG, proactive agents, and end-to-end encrypted P2P collaboration. Private by default, no cloud or API required. Python, Gradio GUI.","archived":false,"fork":false,"pushed_at":"2026-06-29T08:35:33.000Z","size":4053,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-29T09:19:16.386Z","etag":null,"topics":["ai-assistant","end-to-end-encryption","gradio","llama-cpp","local-first","local-llm","offline-first","personal-assistant","privacy","python","rag","self-hosted"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Mr5niper.png","metadata":{"files":{"readme":"docs/README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-11-20T02:39:19.000Z","updated_at":"2026-06-29T08:07:37.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/Mr5niper/Aegis-Synthesis-Architecture","commit_stats":null,"previous_names":["mr5niper/aegis-synthesis-ai-architecture","mr5niper/aegis-synthesis-architecture"],"tags_count":2,"template":false,"template_full_name":null,"purl":"pkg:github/Mr5niper/Aegis-Synthesis-Architecture","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Mr5niper%2FAegis-Synthesis-Architecture","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Mr5niper%2FAegis-Synthesis-Architecture/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Mr5niper%2FAegis-Synthesis-Architecture/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Mr5niper%2FAegis-Synthesis-Architecture/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Mr5niper","download_url":"https://codeload.github.com/Mr5niper/Aegis-Synthesis-Architecture/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Mr5niper%2FAegis-Synthesis-Architecture/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35168795,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-05T02:00:06.290Z","response_time":100,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-assistant","end-to-end-encryption","gradio","llama-cpp","local-first","local-llm","offline-first","personal-assistant","privacy","python","rag","self-hosted"],"created_at":"2025-12-27T11:04:22.693Z","updated_at":"2026-07-05T21:00:29.229Z","avatar_url":"https://github.com/Mr5niper.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Aegis Synthesis\n\n\u003cimg width=\"250\" height=\"250\" alt=\"image\" src=\"https://github.com/user-attachments/assets/777b0949-44b8-4943-bec2-f632390e57d9\" /\u003e\n\nA personal AI assistant that runs entirely on your own Windows PC. No cloud, no\naccount, no API keys. Your conversations and data stay on your machine.\n\nThere are two ways to use Aegis:\n\n- **Just run the app:** download the release zip, extract it, and run the\n  included `Aegis.exe`. No Python or developer tools needed. Start with\n  \"Running the app\" below.\n- **Build or run from source:** developers can build the executable or run from\n  Python. See \"Building from source\" near the end.\n\n## Features\n\n- Local LLM: Llama 3.2 3B or Mistral 7B (GGUF), switchable at runtime. Only the\n  model in use is held in memory.\n- GPU-accelerated: the Windows build uses your GPU (AMD, NVIDIA, or Intel) when\n  present and falls back to the CPU when not, with nothing extra to install.\n- Private by design: conversations and data are stored locally on your PC.\n- RAG knowledge base with semantic search, and a consent-based memory system\n  that asks before saving facts about you.\n- Web search on demand, limited to sites you allow in the settings file.\n- Optional proactive suggestions and secure peer-to-peer collaboration\n  (both off by default).\n- Runs as a single Windows executable, or from source on Windows, macOS, or\n  Linux.\n\n---\n\n# Running the app\n\nThis section is for running the prebuilt `Aegis.exe`. You do not need Python or\nany developer tools.\n\n## What is in the release zip\n\nAfter you extract the release zip you get a folder containing:\n\n- `Aegis.exe` - the application (a single file).\n- `config.yaml` - a settings file that must stay next to `Aegis.exe`.\n- `README.md` - this guide.\n\nKeep `Aegis.exe` and `config.yaml` together in the same folder. They are already\nside by side in the extracted folder; just leave them that way.\n\n## Requirements\n\n- Windows 10 or 11 (64-bit).\n- 16 GB of RAM recommended. Aegis keeps only the model you are using in memory\n  (it loads one at a time), so 8 GB can work, though more is better for the\n  large model.\n- A GPU (AMD, NVIDIA, or Intel) is optional but makes replies much faster. The\n  app uses it automatically when present; nothing beyond your normal graphics\n  driver is needed. With no usable GPU it runs on the CPU.\n- About 10 GB of free disk space. On first run the app downloads two AI models\n  (about 6 GB total) plus a small text-processing model, and stores them next\n  to the exe.\n- An internet connection the first time you run it, to download the models.\n  After that it works offline.\n\n## Setup\n\n1. Extract the zip. Right-click it and choose \"Extract All\", or use your usual\n   unzip tool. Do not run `Aegis.exe` from inside the zip without extracting -\n   it needs `config.yaml` next to it and room to create its folders.\n2. Move the extracted folder somewhere you can write to, for example `C:\\Aegis`.\n   Avoid `C:\\Program Files` and the Desktop; pick a normal folder.\n3. That is it. There is nothing to install.\n\n## Running it\n\nDouble-click `Aegis.exe`.\n\nA black console window opens and stays open while Aegis runs - that is normal,\nleave it open. After a few moments your web browser opens automatically to the\napp at `http://127.0.0.1:7860`. If it does not open on its own, open your browser\nand go to that address manually.\n\nTo quit, close the browser tab and then close the black console window.\n\n### Windows SmartScreen warning\n\nBecause this is a small independent app and is not code-signed, Windows may show\na blue \"Windows protected your PC\" box the first time. If you trust the source of\nthis download, click **More info**, then **Run anyway**. This warning is normal\nfor unsigned software and usually does not appear again.\n\n## First launch: what to expect\n\nThe very first run is the slow one. In order, it will:\n\n1. Download both AI models - the default (about 2 GB) and the larger one\n   (about 4 GB), roughly 6 GB in total. This can take a while depending on your\n   internet speed. The console shows download progress.\n2. Download a small text-processing model (a few seconds to a minute).\n3. Load the models and open your browser to the app.\n\nThe models are saved into a `models` folder created next to `Aegis.exe`, so the\ndownload only happens once. A `data` folder is also created next to the exe to\nstore your conversations and settings. Every later launch starts in well under a\nminute (no downloads, just loading the models).\n\nIf nothing seems to happen for a while on first run, it is almost certainly the\nmodel download in progress - watch the console window for activity.\n\n## Using Aegis\n\n- Type a message in the **Your Message** box and press Enter or click **Send**.\n- The reply streams in as it is written. **Stop** cancels a reply in progress.\n- **Clear** starts a new conversation. **Export** saves the conversation to a\n  file.\n- Rate replies with **Good**, **Bad**, or **Needs Correction** to help it\n  improve over time.\n\nThings to try:\n\n- \"Hello\"\n- \"What is 23 * 456?\"\n- \"What is the latest stable version of Python?\" (it can search the web)\n- \"Remember that my favorite color is blue\" (it asks before saving facts)\n\n### Two models\n\nAegis comes with two AI models, both downloaded on first run, and you can switch\nbetween them in the **Models** panel on the right:\n\n- **default** (Llama 3.2 3B): faster, lighter, good for everyday use.\n- **large** (Mistral 7B): more capable but heavier and slower, and uses more\n  RAM.\n\nBoth models are downloaded on the first run, so switching does not download\nanything. Only one model is loaded at a time: switching unloads the current one\nand loads the other, which takes a few seconds. Your conversation carries over.\n\n## Changing settings\n\nOpen `config.yaml` (next to the exe) in any text editor like Notepad. It controls\nthings such as which web sites the assistant may visit, whether background\nsuggestions are on, and how many reasoning steps it takes. Save the file and\nrestart `Aegis.exe` for changes to take effect.\n\nYou do not need to touch this file to use Aegis; the defaults work out of the box.\n\n## Privacy\n\nEverything runs locally. The AI model runs on your own machine (on your GPU when\nthere is a usable one, otherwise on your CPU), your conversations are stored only\nin the `data` folder on your PC, and nothing is sent to any company.\nThe only time Aegis reaches the internet on its own is:\n\n- the one-time model download on first run, and\n- web searches, but only when you ask a question that needs current information,\n  and only to the sites allowed in `config.yaml`.\n\n## Troubleshooting\n\n**The black console window closed immediately.**\n`config.yaml` is probably not next to `Aegis.exe`, or the folder is not writable.\nMake sure both files are together in a normal folder (not Program Files), and try\nagain.\n\n**The browser did not open.**\nOpen it yourself and go to `http://127.0.0.1:7860`. Make sure the console window\nis still open.\n\n**First run is stuck / taking forever.**\nIt is downloading the model. Watch the console window - if there is activity, let\nit finish. A slow connection can make the first run take several minutes.\n\n**It says it cannot download the model.**\nCheck your internet connection and try again. If your network blocks the\ndownload, you can place the model file manually: create a `models` folder next to\n`Aegis.exe` and put the `.gguf` file there (the exact file name and download link\nare in `config.yaml`).\n\n**Replies are slow.**\nSpeed depends on your hardware. Aegis uses your GPU automatically when it can; on\na CPU-only machine the model runs on the CPU, which is slower. Use the **default**\nmodel rather than **large**, and close other heavy programs.\n\n**Windows blocked it (SmartScreen).**\nSee the SmartScreen note above - More info, then Run anyway.\n\n---\n\n# Building from source\n\nFor developers who want to build the executable or run from Python. Full setup,\nusage, and troubleshooting are in `docs/INSTALL.md`; the short version follows.\n\nRequirements: Python 3.13 (pinned and tested against 3.13.12), and the same disk\nand RAM as above.\n\n## Run from source\n\n```bash\ngit clone https://github.com/Mr5niper/Aegis-Synthesis-Architecture/ aegis-synthesis\ncd aegis-synthesis\n\npython -m venv venv\n# Windows:\nvenv\\Scripts\\activate\n# macOS/Linux:\nsource venv/bin/activate\n\npip install torch --index-url https://download.pytorch.org/whl/cpu\npip install \"llama-cpp-python==0.3.2\" --only-binary :all: --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu\npip install -r requirements.txt\n\npython -m src.main_gui\n```\n\n`torch` and `llama-cpp-python` are not listed in `requirements.txt`; they install\nfrom their own CPU wheel indexes (see the comments in `requirements.txt`). The\nfirst run downloads both models (about 6 GB total) into the `models` folder, then\nopens your browser to `http://127.0.0.1:7860`.\n\n## Build the Windows executable\n\n```bat\nBUILD_EXE.bat\n```\n\nThis creates a clean virtual environment, installs the CPU wheels and pinned\nrequirements, and builds a single `Aegis.exe` into the `dist` folder with\n`config.yaml` copied beside it. The models and data folders are not bundled; they\nare created next to the executable on first run.\n\n## Build on Linux or macOS\n\nPyInstaller cannot cross-compile, so a Linux build must be made on Linux and a\nmacOS build on macOS. See `docs/BUILD_CROSS_PLATFORM.md` for the build script\n(`build_executable.py` / `makefile`) and the per-platform notes. These paths are\nprovided but are not maintainer-tested.\n\n## Documentation\n\n- `docs/INSTALL.md` - full setup, first run, usage, and troubleshooting.\n- `docs/BUILD_CROSS_PLATFORM.md` - Linux/macOS build path.\n- `docs/CHANGELOG.md` - version history.\n- `config.yaml` - configuration reference.\n\n---\n\n## A note on what this is\n\nAegis is an independent, locally-run assistant using small open models. It is not\nas capable as large cloud AI services, and the smaller model can occasionally get\nthings wrong or phrase things oddly. Its strength is privacy and running entirely\non your own machine.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmr5niper%2Faegis-synthesis-architecture","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fmr5niper%2Faegis-synthesis-architecture","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fmr5niper%2Faegis-synthesis-architecture/lists"}