Documented fire_every and skip_first on templated reminders, including the late-conversation reminding pattern for combating context rot and the post-compaction counter reset
Added a Tool Output Transform section covering jq projection of tool responses, with real web_search / web_get shapes and a search β fetch pairing example
Added Glossaries documentation covering term-to-expansion mappings, matching behavior, and attaching glossaries to agent steps via glossary_expansion reminders
Added Pipelines documentation (quickstart, concepts, sources, verification, dead letters) for automated bulk processing of external data through agents
Added Multi-Client Steering documentation covering multi-device session consumption, behavior: steer vs follow_up, and mid-turn interrupts
Added a dedicated Context Engineering section with an opinionated overview and deep-dive pages for Compaction, Reminders, Skills, Tool-Output Offloading, and Steps
Overview covers the pillars of a well-engineered agent: retrieval, instructions, tool configuration, sub-agent delegation, long-session management, tool-output protection, and monitoring
Updated the API Reference spec with all features released JanuaryβApril 2026
Replaced x-supported-api-key-type annotations with x-required-roles on all 133 endpoints, showing exactly which API roles, corpus roles, and agent roles grant access
Each endpoint page now shows role badges in the authorization section
Reorganized API Reference sidebar into grouped sections (Agents, Data, Search and Retrieval, Access Control, Analytics, Platform, Quality and Evaluation)
Rewrote Role-Based Access Control documentation to cover all four role tiers: API roles (13 account-level roles), corpus roles (viewer/editor/administrator/owner), agent roles (agent_user/agent_viewer/agent_developer/agent_administrator), and platform roles
Added API conventions and authentication documentation to the API Reference introduction page
Enhance Agents documentation for Tools and Instructions (#503)
Refresh API Reference (#501)
Introduce new section about Private Deployment options such as on premise (#498)
Enhance documentation around data I/O, Agents, model selection, and BYO-LLM with additional API Concepts cleanup for accuracy and discoverability (#496)