Agent-Friday

KO | EN

gitlite — search

#react #go #android #linux #docker #java #nodejs #kubernetes #vue #pytorch #security #mcp #openai #open-source

(★ 13)

Agent Friday is a desktop AI agent with a holographic user interface that grows new interfaces, skills, and personality traits to suit your needs.

Agent-Friday Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Agent Friday

[![CI](https://github.com/FutureSpeakAI/Agent-Friday/actions/workflows/tests.yml/badge.svg)](https://github.com/FutureSpeakAI/Agent-Friday/actions/workflows/tests.yml)
[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://raw.githubusercontent.com/FutureSpeakAI/Agent-Friday/main/LICENSE)
[![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://python.org)
[![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)

> **Note:** Agent Friday Desktop is the standalone desktop application (this repo). It is distinct from the [Asimov's Mind Claude Code plugin](https://futurespeak.ai/asimovs-mind), which is a separate product built for the Claude Code environment.

---

## What is this?

**Agent Friday** is a privacy-first, self-improving personal AI that runs entirely on your machine. It features a tiered data vault that keeps sensitive information off the cloud, a holographic Three.js interface, a layered content-safety classifier, and a skill-evolution engine — all served by a local Flask app backed by Anthropic Claude, Google Gemini, Ollama, or any OpenAI-compatible provider (OpenRouter first-class, ten providers built in).

Think Jarvis with a sharp newsroom editor's instincts, a sovereign conscience, and a zero-trust data policy.

---

## Demo

[![Agent Friday — Live Demo](https://img.youtube.com/vi/JeAywoHd_jg/maxresdefault.jpg)](https://youtu.be/JeAywoHd_jg)

| | |
|---|---|
| [![Full Explainer](https://img.youtube.com/vi/uFKAQ3uz2U4/hqdefault.jpg)](https://youtu.be/uFKAQ3uz2U4) | [![Defeating Disinformation](https://img.youtube.com/vi/Do2ONuv_UbM/hqdefault.jpg)](https://youtu.be/Do2ONuv_UbM) |
| Full system explainer | Defeating disinformation |

---

## Quick Start

**Easiest (Windows):** download `AgentFriday.exe` from the [latest GitHub release](https://github.com/FutureSpeakAI/Agent-Friday/releases/latest) and run it — no Python required. Or install from source:

```bash
git clone https://github.com/FutureSpeakAI/Agent-Friday.git
cd Agent-Friday
pip install -e .
friday doctor    # verify Python, Ollama, the bundled Gemma model, keys, disk
friday           # launches the server, runs voice-first onboarding, opens http://localhost:3000
```

**No API key? No problem.** Agent Friday ships a bundled local model — **`gemma3:4b`**
(Google's open Gemma 3 4B, runs on ~8 GB RAM). The installers auto-install Ollama and
pull it, so **chat works fully offline with zero cloud keys**. On first run, Friday
greets you by voice and walks you through setup. Cloud keys are *optional upgrades* for
sharper reasoning, image/video generation, and richer voice — add them any time in
Settings (creative/voice degrade gracefully with a clear notice until you do).

> Bigger machine? Upgrade the local brain: `ollama pull gemma3:12b` (or `gemma3:27b`)
> and set it in Settings → Models.

**Adding cloud keys (optional).** The recommended way is `friday setup` — it
stores each key **encrypted** via the credential store (DPAPI/AES-256-GCM),
never in plaintext, never in the repo:

```bash
friday setup        # interactive: keys, model, vault passphrase — all encrypted at rest
```

Or add them any time in the running app under **Settings → Providers**. Both
paths write to the encrypted store. Environment variables (`ANTHROPIC_API_KEY`,
`GEMINI_API_KEY`, `OPENAI_API_KEY`) still work for CI or advanced setups, but a
plaintext key file (e.g. a hand-edited `start.bat`) is **not** the recommended
pattern — prefer `friday setup`.

See [docs/INSTALLATION.md](docs/INSTALLATION.md) for the complete setup guide, including the one-line shell installer, GPU setup, Ollama, and the Windows SmartScreen bypass.

---

## Documentation

| Document | Description |
|----------|-------------|
| [Architecture](docs/ARCHITECTURE.md) | System diagrams, pipeline flows, Mermaid charts |
| [API Reference](docs/API.md) | Every endpoint with methods, paths, request/response |
| [Installation](docs/INSTALLATION.md) | Fresh-machine setup, prerequisites, troubleshooting |
| [Configuration](docs/CONFIGURATION.md) | All `settings.json` options |
| [Skills](docs/SKILLS.md) | Skill system, SkillOpt, auto-research loop |
| [SELF.md](SELF.md) | Friday's self-knowledge document |
| [Credits](CREDITS.md) | Third-party libraries and inspirations |
| [Threat Model](THREAT_MODEL.md) | Security posture, trust boundaries, known gaps |

---

## Key Features

- **Sovereign Vault** — TIER 1/2/3 access control; TIER_2 (private) and TIER_3 (sensitive) data never leave the local model. AES-256-GCM + Argon2id at rest.
- **Layered Safety Classifier** — Fail-closed egress gate with sensitivity classifier as single source of truth; HMAC-SHA256 signed behavioral constraints (Asimov's cLaws).
- **Holographic UI** — Three.js WebGL interface with audio reactivity, process orbs, and personality evolution visualized as progressively complex geometric structures.
- **Knowledge Galaxy** — Your wiki as a navigable 3D galaxy: pages are stars, links and title-mentions are filaments, wiki sections cluster into glowing constellations. Fly through it, hover to trace connections, double-click a star to open the page. Behind it, a two-tier knowledge graph: an always-on structural tier (no LLM, instant, works offline) plus an opt-in GraphRAG semantic tier — **local-only by default**, with sensitive-derived data vault-encrypted at rest and an adversarial egress test suite guarding the cloud boundary.
- **Voice Mode** — Real-time WebSocket audio pipeline; on-device Whisper + Piper by default (Tier-1 CPU, Tier-2 NeMo GPU), or optional Google Gemini Live cloud voice with barge-in interruption and auto-reconnecting hours-long sessions.
- **Universal Tool Loop** — Anthropic, Gemini, Ollama (Gemma native tool calling), and OpenAI-compatible providers share a single agentic tool loop.
- **Creator Economy Layer** — Music (Lyria 3), video (Veo), image generation (Nano Banana Pro/2), provenance (C2PA), federation (Ed25519 identity, X25519+ChaCha20-Poly1305 transport), marketplace, and economy engine.
- **Creation Tools** — Ask in chat or voice for a slide deck (`create_presentation`) or a multi-page website (`create_website`); a deterministic template renders a polished, self-contained HTML artifact into the Studio gallery — keyboard-nav decks with speaker notes and print-to-PDF, responsive sites that deploy anywhere.
- **Content Pipeline** — A full social-media publishing system: compose platform-native posts from your Studio creations (in your voice, with per-platform previews), schedule them on a calendar with learned optimal times, and Friday publishes autonomously via the internal scheduler — LinkedIn, X, Instagram, YouTube, TikTok, Bluesky, Mastodon, Reddit, Substack/Medium (assisted), and the Friday Federation. Every post passes the harm floor and egress gate (hold-for-review, never silent redaction), carries Ed25519 provenance, and feeds local-only analytics that learn what works.
- **Self-Improvement** — Weekly epistemic calibration, SkillOpt nightly loop, closed-loop learning from real usage.
- **Defederation & Moderation** — Asimov-governed defederation protocol, H1–H4 harm floor, community content-policy packs.

---

## Requirements

- Python 3.10+
- **Ollama + the bundled `gemma3:4b` model** — the zero-key default; chat runs fully local ([ollama.com](https://ollama.com), auto-installed by the installers)
- Anthropic API key — *optional upgrade* for live Claude reasoning ([get one](https://console.anthropic.com/settings/keys))
- Google Gemini API key — *optional* for voice and creative features ([get one](https://aistudio.google.com/apikey))

---

## Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md). Issues and pull requests are welcome — please read the [Code of Conduct](CODE_OF_CONDUCT.md) first.

---

## License

Created by **[FutureSpeak.AI](https://futurespeak.ai)** · Built with **Claude by Anthropic** as AI development partner.

Agent-Friday

Agent-Friday Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Agent-Friday

Agent-Friday Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README