AI OS 🟡 BETA
The platform is evolving from a conversational bot server into an autonomous personal operating system. This chapter documents the AI OS capability set: always-on agent workspaces, the tabbed suite shell, adaptive scheduled agents, the skills marketplace, durable memory, real-time voice, consented desktop automation, the hardened browsing control plane and enterprise connectors.
Every capability ships behind a feature flag and preserves legacy behavior
when disabled. Backend entry points are additive namespaces (/api/agent/*,
/api/automations/*, /api/marketplace/*, /api/memory/*, /api/browser/*,
/api/consent/*, /api/connectors/*, /api/v1/sandbox).
In this chapter
- Agent Mode & Snapshots
- Workspace Tabs
- Adaptive Automations
- Skills Marketplace
- Memory OS
- Consent System
- Unified Settings
Feature flags (botserver Cargo features)
| Feature | Crate | Surface |
|---|---|---|
agent-vm | botagent | per-session VMs, snapshots, sandbox exec, org API keys |
automations | botautomation | NL schedules, run engine, delivery loop, dashboard API |
marketplace | botmarketplace | public catalog, publish/install |
consent | botconsent | permission matrix, prompt cards, audit |
memory-os | botmemory | recall injection, extraction hook, memory API |
connectors | botconnectors | third-party ingestion with ACL filtering |
browser-policy | botbrowserpolicy | browsing policy, budgets, page facts |
channel-bindings | botchannelbindings | default number/domain bindings per bot |
Model routing (multi-profile fallback, circuit breakers, council mode) lives in
the botllm crate and activates when GB_LLM_PROFILES (env or JSON file) is
present; single-provider deployments continue to work unchanged.