One desktop app for every LLM, with a strict no-uploads policy

What it does

Askimo is a Kotlin-based desktop application that consolidates chat, retrieval-augmented generation, autonomous agent delegation, and MCP tool integration into a single native client. It talks to a broad roster of LLM providers—OpenAI, Anthropic, Gemini, Grok, Ollama, and many others—while storing conversations, RAG indexes, and telemetry in a local SQLite database. The pitch is simple: stop juggling browser tabs and stop worrying about your files leaking to a SaaS backend.

The interesting bit

The project treats privacy as a product feature rather than an afterthought. Everything from the hybrid BM25 + vector RAG engine to the usage/cost telemetry runs offline by default, and the app even bundles local speech-to-text and text-to-speech options so you can dictate messages without touching a cloud API. That’s unusual in a space crowded with Electron wrappers and browser-based clients.

Key highlights

• Native desktop app built with Compose Multiplatform, not a browser wrapper; the README claims a lean memory footprint of 50–300 MB for the app itself.
• Supports a wide provider roster: OpenAI, Anthropic, Gemini, Grok, Ollama, LM Studio, OpenRouter, NVIDIA NIM, vLLM, and generic OpenAI-compatible endpoints.
• Local RAG with hybrid BM25 and vector search, plus an AI classifier that skips retrieval when the query doesn’t need it.
• Reusable Skills defined once and materialized into any supported agent CLI (Claude Code, Codex, Antigravity).
• Multi-step Plans engine for chaining AI workflows from a form UI, with live progress and export to PDF or Word.
• Local/offline voice mode via Whisper-compatible and Piper endpoints.

Caveats

• Agent CLI support assumes you have already installed Claude Code, Codex, or Antigravity yourself; Askimo streams their output but does not bundle them.
• The README highlights RAG and Plans features without detailing implementation specifics like embedding models or index size limits.
• AGPLv3 licensed, which will deter some commercial adopters.

Verdict

Developers who want a single, privacy-minded desktop client for juggling multiple LLMs and local file search should take a look. If you are looking for a headless API server or a fully embedded agent framework, this is not it.