Agentica v0.2.3

Distribution repo for Agentica binaries

Agentica v0.2.2 — voice that actually works

Local voice fixed end-to-end (Silero VAD v5 + correct packaged asset paths), Whisper+Kokoro bundled. Remote SSH/Ollama now lists the box's real installed models + one-click remote download. Model picker shows installed vs not-downloaded. Remote expressive (Chatterbox) + Gemini voice settable in-app. SLURM jobs verified on Discovery. Security hardening from the v0.2.x review.

macOS arm64 + Linux x64. (Intel mac build pending a CI runner.)

Agentica v0.2.0 — Interactive voice mode

Agentica v0.2.0 — Interactive voice mode + a visual canvas

Talk to Agentica hands-free, and watch it answer with live visuals.

🎙️ Interactive voice mode

• Continuous conversation — no tap-to-talk. Always-listening (Silero VAD), and you can interrupt it mid-sentence (barge-in).
• Open-weight + local — Whisper (speech-in) + Kokoro (natural speech-out), all on your machine. Models download on first use.
• Cloud option — Gemini Live realtime, opt-in with your own API key (Settings). The agent + canvas still run locally.
• Expressive remote voice — point it at a GPU box for a heavier voice (Chatterbox/Dia/XTTS) over an SSH tunnel.

🖼️ A canvas beside the conversation

• Ask "what's the weather" and get a multi-day forecast chart; the agent also renders charts, tables, markdown, and images.
• Embedded live web pages you can click, and an agent-driven browser whose screenshots stream into the canvas.
• Click a card to keep the conversation going.

Under the hood

• Agentic chat now streams tool steps + visuals mid-turn (was all-at-once).
• New backend voice gateway, canvas tools, and provisioning — all open-weight.

Note: the macOS build is larger (~230MB) because the natural-voice runtime ships with it. First voice use downloads the voice models.

Agentica v0.1.2

Agentica v0.1.2 — first release with macOS binaries, and it just works.

curl -fsSL https://agenticaai.vercel.app/install.sh | bash    # macOS (Apple Silicon) · Linux · WSL

What's new

• macOS support — ships Apple Silicon (arm64) .dmg + .zip with a working bundled engine (the first release to include Mac builds).
• Zero setup — on first run the app auto-provisions Ollama (no manual install, no sudo) and pulls a small default model (qwen3.5:4b-mlx) — unless you already have a model, which it reuses. Nothing to click.
• Chat fixed — agentic chat no longer fails with a read-only-filesystem error; answers are fast (no minutes-long chain-of-thought); the model now knows the current date/time and which OS it's on.
• Open-and-it-works — the app retries the local engine on launch (a calm "Starting…" instead of an error) and auto-reconnects.
• Chat polish — sticky auto-follow with a "↓ Latest" button; live "thinking" for reasoning models; cancel running/queued remote jobs; SLURM clusters selectable as job targets.
• Linux — x64 AppImage + tarball, each with a working bundled engine.

Intel macOS (x64) and additional Linux variants will follow via CI.

Agentica v0.1.0

First public release. Install with:

curl -fsSL https://agenticaai.vercel.app/install.sh | bash

Linux (x64) builds are attached. macOS (.dmg, Apple Silicon + Intel) builds are added by the release CI on the next tagged build. Uninstall anytime with agentica-uninstall.