Understand
Multilingual semantic scoring extracts the goal, formats, and privacy signals.
FULLY LOCAL · EXPLAINABLE · USER-CONTROLLED
Describe the outcome; build an auditable workflow across ByteQuant tools
Local Agent analyzes a natural-language goal, finds suitable tools, and creates a multi-step plan. You approve and run every step; data and the decision summary stay in this tab.
✓ Synced with 89 tools✓ 0 remote model calls✓ Visible decision summary
Try local planning →BQ-Agent 1.0Local hybrid decision model
Versioned semantic scoring plus explainable planning rules. No network request, remote model, or hidden chain-of-thought.
LOCAL-FIRST ORCHESTRATION
One intelligent layer across every existing tool
Core tool behavior is unchanged. The agent coordinates discovery, planning, and only the data handoff you explicitly approve.
Plan
Versioned rules rank suitable tools and disclose the reason for each choice.
Run
Open each step and pass input or output inside this tab only when you act.
Verify
The error translator explains a likely cause, safe checks, and the method boundary.
PRIVACY BY ARCHITECTURE
A genuinely in-browser architecture
bytequant.org · bytequant.org
- ✓
Neither prompts nor tool inputs go to a ByteQuant server or third-party model.
- ✓
The plan uses sessionStorage for the open tab only; content is never written to localStorage.
- ✓
Voice input activates only after the browser verifies on-device processing; there is no remote fallback.
- ✓
This is not a generative foundation model. It is an explainable hybrid search and planning engine, not legal, security, or identity verification.
FAQ
About Local Agent
Is Local Agent an LLM?+
No. It is not a generative LLM with hidden weights or remote inference. It uses multilingual semantic matching and explainable, versioned planning rules.
Can it run tools without me?+
No. File selection, tool execution, downloads, and step transitions require user action.
Is voice sent anywhere?+
No. The voice control starts only if the browser verifies on-device speech recognition; otherwise it captures nothing.
Is the decision summary chain-of-thought?+
No. Hidden internal reasoning is not exposed. The summary lists matched signals, extracted parameters, tool-choice reasons, and limitations.