Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.protean.peerislands.io/llms.txt

Use this file to discover all available pages before exploring further.

Winter is the parent brand. The Protean is the product Winter ships: a runtime, AI foundations, and a developer system. It runs on your laptop and in the cloud from one codebase.

​
What it is

A platform for AI-native systems. Four execution principles drive every design decision:
  1. Local-first — every capability runs offline on a developer’s machine before it runs in cloud.
  2. Deterministic — same input + same seed + same prompt cassette = same trace.
  3. Replayable — runs persist their event log; can be replayed verbatim for debugging, testing, eval.
  4. Streaming-first — SSE/WS from the runtime up through SDK, UI, and CLI.
See execution-philosophy for the long-form version.

​
What’s in the box

  • Runtime (@protean/runtime) — Run entity, deterministic clock + RNG, executor abstractions.
  • LLM gateway (@protean/llm) — wraps pi-ai, multi-provider via OpenRouter, cassette mode for replay.
  • Foundations — Council, Watchers, Insights, DeepResearch (one per slice; Slice 1 ships Company Research).
  • API (apps/api) — Hono on Bun. OpenAPI 3.1 from Zod schemas.
  • Web (apps/web) — Vite + React shell. Imports @protean/sdk only.
  • CLI (apps/cli) — protean shim around @mariozechner/pi-coding-agent.
  • Desktop (apps/desktop) — Electron wrapper around the web build.

​
What it isn’t

We deliberately don’t build:
  • A general-purpose agent framework (use LangChain or pi-mono directly).
  • A LLM-as-orchestrator product (Run is the orchestrator; LLMs are tools).
  • A no-code builder.
See SOUL.md in the repo for non-negotiables.