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

Feature flags (botserver Cargo features)

FeatureCrateSurface
agent-vmbotagentper-session VMs, snapshots, sandbox exec, org API keys
automationsbotautomationNL schedules, run engine, delivery loop, dashboard API
marketplacebotmarketplacepublic catalog, publish/install
consentbotconsentpermission matrix, prompt cards, audit
memory-osbotmemoryrecall injection, extraction hook, memory API
connectorsbotconnectorsthird-party ingestion with ACL filtering
browser-policybotbrowserpolicybrowsing policy, budgets, page facts
channel-bindingsbotchannelbindingsdefault 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.