Hero Background
The Edge-Computed Web Agent

Forge an agent with a soul.

Like the mythological centaur, Chiron binds human intuition with raw autonomous horsepower. It's not just an agent—it's an extension of your own will that learns while you sleep.

The Centaur Relationship

Fully autonomous bots break. Human-only workflows are slow. Chiron represents the ideal synthesis: you provide the high-level goal and human intuition, and Chiron provides the tireless execution.

  • Human-in-the-Loop (HITL): Chiron pauses when unsure, allowing you to approve actions or correct its course.
  • Shared Context: Operates directly in your browser, inheriting your active logins and sessions natively.

Architected for the Edge

Chiron redefines what a browser extension can do. Summon it from any tab using Alt+Space, or right-click any text to dispatch the agent instantly.

Smart Model Orchestration

Chiron routes high-complexity DOM tasks to flagship models (GPT-4, Claude), while seamlessly falling back to blazing-fast on-device LLMs for utility tasks, drastically cutting your API costs.

Self-Healing Workflows

When websites update and break crystallized macros, Chiron detects the failure, hands control back to the AI to visually solve the new layout, and writes a new macro automatically.

Multi-Modal Perception

By seamlessly stitching together cross-origin iFrames, Accessibility Object Models (AOM), and viewport screenshots, Chiron is able to "see" the web exactly as you do.

Absolute Serverless Privacy

There are no middlemen. No backend databases. No telemetry. Chiron connects directly from your Chrome extension to the AI provider of your choice locally or via public API. Your data, memories, and vector embeddings are stored entirely in your local browser's IndexedDB.

Inside the Extension

A seamless side panel that lives alongside your work,
ready to assist or take over.

1 The Omnibar

Seamlessly toggle between conversational chat and autonomous agent mode. Give Chiron a task and watch the execution trace unfold.

2 The Forge (RAG)

Drag and drop documents directly into the extension. Chiron chunks and embeds them locally using WebGPU, expanding its knowledge instantly.

3 Macro Library

View and play back crystallized macros. If the agent solves the same task twice, it saves the exact UI steps for zero-shot, instant execution later.

4 Night Shift

When you step away, Chiron consolidates its memories, extracts negative/positive SOPs, and evolves the Anima's core persona.

5 Edge-Computed Execution

Run powerful models like Llama 3.2 or DeepSeek-R1 completely locally via WebGPU, or connect your own API keys for frontier cloud tasks.

6 Anima Portability

Agents are fully portable. Export your Anima's persona, memories, and learned macros as a single file to share with your team or deploy elsewhere.

Chiron Logo
Work
24.5k
User: Extract the invoice totals from these PDFs and upload them to Xero.
Plan (Sub-goals): - Query The Forge for PDFs - Extract vendor, date, and total - Bulk upload to Xero Expenses
[Reasoning] I need to search the local knowledge base for the Q3 invoices. [Actions] QUERY_FORGE "Q3 Invoices folder"
[Reasoning] Parsing 4 PDF documents for vendor, date, and total. [Actions] EXTRACT_SCHEMA [vendor, date, total]
Approval Needed! Reasoning: Ready to draft expense entries in Xero.

Preferences

The Forge

Create and manage Anima profiles.

Drag & Drop files here

Supports TXT, MD, PDF, CSV

Q3_Invoices.pdf
Vendor_List.csv

Preferences

The Forge

Create and manage Anima profiles.

Upload to Xero
Enrich CRM Lead

Night Shift Active

Consolidating memories & evolving persona...

> Processing 14 recent tasks...

> Extracting semantic anchors...

> Synthesizing SOP for "Xero Uploads"...

> Crystallizing 1 new macro...

> Persona evolution complete_

Settings

Execution Engine

Configure how Chiron processes tasks.

100% Local (WebGPU)
Llama-3.2-1B-Instruct
Live Generation Speed 35.4 t/s

Preferences

Anima Portability

Export your agent's soul (persona, memories, macros) to a portable file.

Data Management

Backup or restore your entire Chiron ecosystem.

Agents with a Soul

Meet the Anima—an encapsulated, exportable intelligence module. Animas aren't just prompts; they are evolving digital entities with procedural memory.

The Forge (RAG)

Drag and drop PDFs, Markdown, and text files. The Anima absorbs them using 100% local edge-computed vector embeddings, expanding its knowledge base instantly.

Night Shift

When you step away, Chiron gets to work. It autonomously reflects on recent tasks, synthesizes feedback, and evolves the Anima's core persona to better match your preferences.

Macro Crystallization

If Chiron solves the same task repeatedly, it "crystallizes" the workflow into a deterministic macro. Next time, it executes instantly—zero AI overhead required.

A Persona for Every Workflow

Start with a blank slate, or import one of our pre-configured Animas designed for specialized tasks.

The Researcher

Focus: Deep-dive analysis
.anima
"You are a rigorous academic researcher. Cross-reference all claims. Prioritize primary sources..."
12 RAG Documents 4 Learned Macros

The Data Clerk

Focus: Extraction & Entry
.anima
"You are a strict data entry agent. Extract tables into structured JSON. Do not hallucinate fields..."
2 RAG Documents 18 Learned Macros

The Deal Hunter

Focus: E-commerce Logistics
.anima
"Find the absolute lowest price for the requested item. Factor in shipping costs and read reviews..."
0 RAG Documents 9 Learned Macros

The QA Engineer

Focus: End-to-End Testing
.anima
"Execute the provided test suite. If an assertion fails, immediately write the error to the scratchpad..."
5 RAG Documents 22 Learned Macros

Hardware Guidelines

Chiron's edge-computed WebGPU models run entirely on your device. Ensure your hardware matches your ambition.

Cloud-Only / Nano

Any modern device

Offload processing to OpenAI, Anthropic, or Google APIs, or use Chrome's built-in Nano model.

  • • PC / Mac: Minimum 4GB RAM
  • • Chrome 130+
  • • BYOK API Keys required
RECOMMENDED

Standard Local Edge

Llama 3.2 1B / DeepSeek 1.5B

Fast, capable local execution for daily tasks, RAG, and Night Shift persona evolution.

  • • PC: 16GB+ RAM, NVIDIA RTX 3050+ or AMD RX 6600+ (4GB+ VRAM)
  • • Mac: Apple M-Series (8GB+ Unified Memory)
  • • 2GB Free Disk Space

Enthusiast Local Edge

Qwen 2.5 3B

Maximum offline reasoning and macro crystallization. Requires serious hardware to maintain smooth execution.

  • • PC: 32GB+ RAM, NVIDIA RTX 3080/4070+ or AMD RX 7800+ (8GB+ VRAM)
  • • Mac: Apple M-Series Pro/Max (16GB+ Unified Memory)
  • • 4GB Free Disk Space

Zero Servers. Total Privacy.

Chiron lives entirely in your browser. With support for on-device LLMs and BYOK capabilities for frontier models, your data never leaves your control unless you explicitly send it.