ECAssistant.Console 1.0.12

ECAssistant Console

ECAssistant in your terminal, five minutes from now. The assistant-first CLI: a local-first AI agent with tools, memory, sub-agents, and vision — running entirely on your machine. No cloud, no API keys.

We named it ECAssistant because we believe AI is there to assist people — and the Console is that promise in its most direct form: a colleague in your terminal that reads files, runs shell commands, edits code, and shows you everything it does. Every tool call is permission-gated (approve once / always this session / deny). You stay in the loop, always.

Part of ECAssistant — your models, your keys, your machine. MIT.

Install

Public on nuget.org — no token, no auth:

dotnet tool install -g ECAssistant.Console
ecassistant

(Or clone and dotnet run for a source checkout.)

First run

The setup wizard walks you through everything:

1. Choose local or remote — local GGUF models, or any OpenAI-compatible endpoint. Only what you pick is ever downloaded.
2. Model selection — pick from the built-in catalog (chat, vision, embedding models), downloaded with SHA-256 verification.
3. LLM server install (local mode) — server binaries come from the NuGet package; the wizard stages them to ~/.ECAssistantLLM/server/.
4. Backend runtimes — ternary models (e.g. Bonsai/Qwen3.8-27B) get their required llama.cpp runtime installed automatically.

After setup, nothing is downloaded at chat time — it's a finished, offline-capable product.

What you get

• 🧠 Tools — files, shell, git, dotnet, code editing, sub-agents, vision structure. All permission-gated.
• 🖥️ Terminal UI — streaming chat, session tabs, tool-call rendering (powered by ECAssistant.TUI).
• 👁️ Vision — images and PDFs to structured JSON — screenshots, UI mockups, or scanned documents become a fixed, versioned JSON schema (elements, bounding boxes, label↔input associations, semantic groups) — analyzed locally, ready for programmatic use.
• 🤝 AskUser checkpoints — when the model is genuinely unsure it asks you a real question with options instead of guessing.
• 🔒 Local-first — everything runs on your machine; nothing phones home.

Configuration

• appsettings.json — app-level settings (written by the wizard)
• ~/.ECAssistantLLM/llm-server.json — LLM server models and endpoints (managed by the wizard)

Build your own host

The Console is deliberately thin — ~100 lines of wiring around ECAssistantCore. Read ConsoleApplication.cs as the reference for embedding an agent in your app.

The ecosystem

License

MIT — © 2026 SideDevEC