Supporting Infrastructure
README · Agents & Applications · Personal Agents · Projects & Products · Streaming Proactivity
Use this guide to identify the component that supplies context, preserves state, wakes an agent, or serves an interaction model. A working component establishes that capability; intervention quality belongs to the system and policy using it.
| Layer | Representative resources | Role in the system | Inspect next |
|---|---|---|---|
| Observation and context capture | Screenpipe | Screen/audio history and context for downstream agents. | Input scope, timestamps, retention, and how context reaches the intervention policy. |
| Persistent memory | memU | Background memory formation and retrieval across sessions. | Which memories affect a decision, how they are updated, and who decides to surface them. |
| Activation and goal continuity | OpenClaw, Proactivity SDK, Hermes Agent, Letta Code | Heartbeats, event/schedule activation, persistent state, and goal-driven work. | Separate wake-up from the later decision to notify, ask, act, or remain silent. |
| Camera and inference interface | Live VLM WebUI | Connect live camera input to inference backends. | Native streaming state, causal input exposure, and perception during generation depend on the backend. |
Canonical project descriptions, access/licensing boundaries, and checked dates are maintained in PROJECTS.md, with activation runtimes in its runtime section. This page provides an architectural route without duplicating the release registry.
For model-owned memory and response gates, use STREAMING.md. For learning when memory should cause assistance, start with Remember When It Matters and Cognifold. Read their evidence cards before treating a memory improvement as an improvement in user-facing initiative.