Architecture Overview

Daedalus is built as three composable layers, each independently pluggable.

Layer 1: Intelligence Core

The brain. Everything that makes Daedalus personal and adaptive.

Knowledge Management — Three-tier system (critical, important, reference) with semantic search via embeddings
Progressive Learning — Lessons captured from every session, star-rated priority system
OpenSpec Engine — Markdown-based specifications, searchable, creatable from any surface
Multi-Model Router — Local models for simple tasks, cloud for complex. Config-driven, cost-transparent
Adaptive Context — Loads the right amount of knowledge based on the model’s capability

The hands. How work gets done.

Hierarchical Agents — Supervisor → project → task agents with scoped permissions
Line Cook — Primary execution framework (prep → cook → serve → tidy)
Framework Composability — Plug in Strands, Goose, LangGraph, or bring your own
Execution Adapter — Standard interface (6 methods) that any framework implements

The face. How users interact.

PWA-First — One web frontend serves browser, desktop (Tauri), and mobile
Adaptive Modes — Developer mode (terminal, code), knowledge worker mode (chat, docs), minimal mode (just chat)
Communication Gateways — Slack, email, and other channels (bi-directional)
API Foundation — Every surface is a client of the same Go API

Layer	Language	Why
Platform	Go	Always-on, single binary, native concurrency
Intelligence	Python	AI ecosystem (embeddings, provider SDKs)
Desktop	Rust (Tauri)	Native performance, small binary
Frontend	Web	One codebase for all surfaces

Communication between Go and Python uses gRPC, which also serves as the native MCP transport.