Skip to main content

Platform Architecture

Mysta is engineered as an agentic browser stack that mirrors how teams already work: sense information, structure it, act with judgement, and report back. Each layer is modular so we can deploy on a single laptop or scale across a global network.

1. Perception Layer — 输入接口​

  • Captures live DOM state, screenshots, transcripts, and API responses directly from the browser runtime.
  • Normalises raw data into typed events that the orchestrator understands.
  • Supports multimodal expansion (voice, vision, documents) to power richer experiences on the roadmap.

2. Distribution Layer — 分发层​

  • Routes perceived signals into specialised queues (memory ingestion, workflow triggers, supervisor alerts).
  • Applies policy envelopes—allowed domains, rate limits, approval requirements—before an action is even planned.
  • Keeps short-term context close to the tab for low-latency reasoning.

3. Knowledge Fabric — 语料层 & 记忆​

  • Synchronises with the Mysta Memory Server for long-term recall, credential vaulting, and compliance logging.
  • Builds retrievable corpora per team, project, or client so the agent can cite previous runs and artefacts.
  • Enables semantic search and entity tracking to ground decisions.

4. Agent Assembly — Agent 层​

  • Determines the next best action via the orchestrator, blending deterministic flows with LLM planning.
  • Invokes plugins (form filler, scraper, publisher, verifier) as composable skills.
  • Supports distributed execution so tasks can shift between local machines and cloud nodes.

5. Output Interfaces — 输出接口​

  • Surfaces progress inside the Supervisor dashboard, extension panels, APIs, or downstream systems.
  • Produces human-readable reports, structured payloads, and audit trails.
  • Closes the feedback loop by writing outcomes back into memory and notifying stakeholders.

Together these layers convert the browser into an intelligent operations system—ready for enterprise scale yet deployable in minutes.