Your private intelligence.
Arca v0.3.5 adds a real model runtime, memory and improved document reasoning.
Documents
Upload PDF, DOCX, TXT or Markdown. Arca retrieves relevant passages and gives them to the brain.
Documents are parsed locally in this browser.
Ask your documents
No active document.
Arca Core
This is the brain layer: model runtime, context, memory, retrieval and engine routing.
Browser Brain
Open-source LLM running on your device with WebGPU. Model cache: IndexedDB.
Local model
Core routing
The model interprets meaning instead of matching exact words.
CoreRecent conversation turns are retained and supplied to the model.
CoreRelevant chunks are retrieved before the model answers.
CorePrepared for WordPress, server and XHM actions later.
NextVoice
Speak to Arca, then send the transcription through the same Core.
Tap to speak
Uses browser speech recognition when available.
Connections
Arca Core can switch model providers without changing the rest of Arca.
Runs an open-source model directly in this browser.
CheckingOpenAI-compatible local endpoint for installed QVAC.
OptionalOptional fallback provider for harder tasks.
Not configuredLocal parsing, chunking and retrieval.
ReadyDiagnostics and approved actions through Core tools.
NextHealth, logs and diagnostics through Core tools.
NextSettings
Configure optional QVAC and cloud fallbacks. The Browser Brain requires no API key.
QVAC/local endpoint
QVAC model alias
Cloud API base URL
Cloud model
Cloud API key (session only)
Document context characters
Memory turns