Drop Claude Code auto-memory layer; it does not apply to Kai Claude Code's auto-memory loader (~/.claude/projects/.../memory/MEMORY.md) only fires in interactive TTY mode. Kai runs the inner Claude in stream-json mode (--print --input-format stream-json), which bypasses that loader entirely. The previous wiki listed it as Layer 1 of Kai's memory architecture, which was misleading - that layer never runs for Kai's subprocess and only applies to Outer Claude editing this code. Renumber the remaining three layers (Home memory, Mem0/Qdrant semantic, History) and remove the auto-memory step from the Memory flow diagram. Add a top aside and a flow-diagram footnote explaining that Claude Code's auto-memory exists but is intentionally bypassed, so contributors who know about it understand why it is absent.

zigguratt committed Apr 20, 2026

Document /memory commands and per-user MEMORY.md scoping Slash-Commands.md: add a Memory commands section covering /memory, /memory search, /memory stats, /memory forget, and /memory help, with inline examples mirroring the actual bot output. Add corresponding rows to the Quick reference table. System-Architecture.md: split the Memory system from three layers to four. Layer 2 (Home memory) is now per-user at DATA_DIR/memory/<chat_id>/MEMORY.md (per PR #348). Layer 3 (new) is the Mem0/Qdrant semantic memory with explanation of physical-vs-logical isolation. Layer 4 is the renumbered conversation history. The first- message context bullets and the Memory flow diagram are updated to match, with a note that semantic memory hits run on every message rather than first-message-only. Workspaces.md: update the injection table to reflect per-user MEMORY.md and add a row for semantic memory hits.

zigguratt committed Apr 20, 2026

Document context window and autocompact configuration (issue #213) Add CLAUDE_MAX_CONTEXT_WINDOW and CLAUDE_AUTOCOMPACT_PCT to: - Protected Installation: new rows in the Claude config table - System Architecture: subprocess resource configuration details - Troubleshooting: guidance for high token usage / daily limit issues

zigguratt committed Apr 2, 2026

Update wiki for PRs #136-#188 - System Architecture: update file storage paths (DATA_DIR/files/<chat_id>/), memory location (DATA_DIR/memory/MEMORY.md), per-user history directories, file cleanup background task, save prompt before shutdown, health monitor failure tracking, GitHub notification routing, SSRF protection, response size cap, error detail redaction, cooldown pruning - Protected Installation: workspace/ to home/ rename, add history/memory/files per-user subdirectories to directory layout, note runtime data separation - External Services: add allow_path_suffix field, SSRF protection, 10MB response cap - Multi-User Setup: update history and file isolation to use DATA_DIR paths - Workspaces: workspace/ to home/ in home workspace description - Testing: update test count (827 to 1173) and file count (21 to 25), add pool, prompt_utils, telegram_utils, phase2_isolation, user_config, workspace_config test entries - Troubleshooting: add health monitor failure notification section

zigguratt committed Mar 26, 2026

Clarify why API info is injected in all workspaces

zigguratt committed Mar 22, 2026

Add Multi-User Setup page and update existing pages for multi-user support New page: Multi-User-Setup.md covering users.yaml config, per-user isolation, subprocess pool, roles, process isolation, and migration. Updated pages: - System-Architecture: diagram shows pool.py, pool.py in source files, auth mentions users.yaml, database schema notes per-user namespacing, user separation section covers per-user os_user - Home: opening paragraph plural, Multi-User link in setup guides - Getting-Started: auth mentions users.yaml, multi-user in next steps - Protected-Installation: users.yaml in directory layout, user separation section covers per-user os_user vs CLAUDE_USER

zigguratt committed Mar 22, 2026

Document per-workspace configuration (workspaces.yaml)

zigguratt committed Mar 17, 2026

Update System Architecture: add agent subprocess pattern, missing modules, and endpoints

zigguratt committed Mar 13, 2026

Add Testing wiki page and security model section

zigguratt committed Mar 8, 2026

Add deployment modes, user separation, and install.py to System Architecture

zigguratt committed Mar 8, 2026

Add totp.py to source files table

zigguratt committed Mar 3, 2026

Add TOTP Authentication wiki page and cross-references New page documenting setup, daily use, configuration, CLI commands, security model, and cross-platform notes. Cross-references added to Home (setup guides), Getting Started (next steps), Troubleshooting (new TOTP section), and System Architecture (auth step).

zigguratt committed Mar 2, 2026

Document dual-mode Telegram transport (webhook + polling) Update 5 wiki pages to reflect the new dual-mode transport support. Polling is the zero-config default; webhook mode requires a tunnel.

zigguratt committed Feb 25, 2026

Document ALLOWED_WORKSPACES config feature

zigguratt committed Feb 24, 2026

Document file exchange feature and add missing CRUD endpoints

zigguratt committed Feb 24, 2026

Add External Services wiki page, fix services.json typo in System Architecture

zigguratt committed Feb 17, 2026

Update history description: scans back as needed, not just 2 days

zigguratt committed Feb 17, 2026

Add service proxy to System Architecture, improve overview diagram

zigguratt committed Feb 17, 2026

Rename Architecture page to System Architecture

zigguratt committed Feb 11, 2026