- Criado serviço systemd para forgejo-runner como gbuser
- Removido runner root antigo
- Todas permissões de /opt/gbo/work agora são gbuser:gbuser
- Fim dos erros de permissão no cache do sccache
- Fixes 'rustup could not choose a version of cargo' error
- Sets stable toolchain as default before cargo build
- Preserves sccache configuration and workspace cache
- Remove configuração automática do ~/.cargo/config.toml
- O workspace já tem .cargo/config.toml que sobrescreve o global
- Devs só precisam rodar ./DEV-DEPENDENCIES.sh para instalar mold/lld/clang
- Config do projeto (./.cargo/config.toml) é usado automaticamente
- Detecta automaticamente o home do usuário atual
- Substitui /home/gbuser hardcoded por $USER_HOME
- Funciona em qualquer máquina (dev local, CI, etc)
- Adiciona .cargo/config.toml com configuração de linker
- Usa clang + lld para linkagem 30-40% mais rápida
- Adiciona sccache para cache de compilação
- Remove comentários do workflow CI
- Ganho total esperado: 17min → 2-5min (incremental)
- Adiciona 'export RUSTC_WRAPPER=sccache' no passo de build
- Adiciona sccache --show-stats antes e depois do build para monitorar cache
- Workflow agora usa sccache corretamente para builds incrementais
- Ganho esperado: 17min → 2-5min (após 1º build)
- Adiciona função configure_cargo() que cria ~/.cargo/config.toml
- Configura linker clang + lld para linkagem rápida
- Reduz tempo de link em ~30-40%
- Aplica para todas as máquinas de desenvolvimento
- Corrigir RUSTC_WRAPPER para usar sccache corretamente
- Corrigir caminho do binary para /opt/gbo/work/target/debug
- Usar hostname 'system' ao invés de IP
- Adicionar comentários sobre features padrão
- Remover killall redundante (systemctl já para)
- Manter features padrão otimizadas (sem ooxmlsdk)
Features padrão já são eficientes:
- chat, automation, cache, llm, vectordb, crawler, drive, directory, kb-extraction
- ooxmlsdk NÃO incluído (apenas em docs feature opcional)
- Economia com sccache: 17min → 2-5min (incremental)
- Combined botserver and botui builds in single workflow
- Removed bottest from CI (no test execution)
- Deploy both binaries to STAGE-GBO system container via SSH
- Simplified to single workflow file
- Removed separate bottest.yaml and botui.yaml
- Temporary fix: Hardcoded SYSTEM_HOST to 10.157.134.196
- Issue: Repository variables not being resolved in workflow
- Next: Test deploy works, then fix variable system
Previously start.bas was prevented from re-running via a Redis flag
(set for 24h). This prevented switchers from being re-sent when
users reloaded the page. Now start.bas executes unconditionally
on every WebSocket connection and every first message.
The switchers were stored in Redis with the DB session ID (session.id)
but retrieved using the WebSocket session ID, causing switchers to
never be found. Fix by saving session.id before the move into
spawn_blocking closure.
- Fix DriveMonitor skipping .bas files that exist but have indexed=false
and same ETag (needs_reindex was computed but not used for bas sync)
- Mark bas files as indexed after successful sync to work dir
- Update INFRA.md: MinIO connection setup (port 9100, creds from drive.service)
- Remove erroneously created bottemplates start.bas (bots live in MinIO)
- Add start.bas with ADD SWITCHER keywords (tables, list, cards)
- Remove /opt/gbo/data references from AGENTS.md and INFRA.md
- Bots are stored in MinIO drive, not in /opt/gbo/data
- Update infrastructure paths documentation
- Fixed KB indexing logic that skipped re-index when DB showed docs but Qdrant was empty
- Added Qdrant collection validation before skipping indexing
- Updated AGENTS.md with correct log locations for staging/production
- Deployed to staging, awaiting CI completion
- Next: monitor chat.stage for KB search functionality
Continuation instructions:
1. Check CI status on ALM (action_run table in PROD-ALM DB)
2. Verify botserver binary updated on staging system container
3. Test KB search: ask question about PDF content in bot
4. Check /opt/gbo/logs/out.log for DriveMonitor indexing activity
5. Verify Qdrant collection has indexed_vectors_count > 0
Root cause: handle_gbkb_change() only checked DB document_count, not Qdrant state
Fix: Added get_collection_info() call to validate Qdrant has points before skipping
- Check Qdrant collection exists and has points before skipping re-index
- Previously only checked database document_count, causing issues when:
* Qdrant collection was deleted but DB still showed docs
* Vectors were not properly indexed (indexed_vectors_count = 0)
* New PDFs added to existing KB folder
- Now validates both DB and Qdrant state before deciding to skip indexing
- Logs warning when DB and Qdrant are out of sync
Fixes issue where KB from PDF files in drive were not being indexed to Qdrant
- Remove 'ADD SWITCHER' → 'ADD_SWITCHER' early replacement in preprocess_basic
which prevented convert_multiword_keywords from properly handling the AS keyword
- Add ADD_SWITCHER (underscore) pattern alongside ADD SWITCHER (space) pattern
so both forms are correctly converted to Rhai custom syntax
- Fix skip condition: only skip ADD_SWITCHER lines when 'as' is lowercase
(already properly converted), not when AS is uppercase (needs conversion)
- Fix ADD\s+MEMBER double-escaped regex → ADD_MEMBER + ADD\s+MEMBER patterns
- Add CLEAR SWITCHERS pattern to convert_multiword_keywords
- Add CLEAR_SWITCHERS() skip condition
Closes#495
- Add is_public column to bots table (migration 6.3.2-01)
- Add user_organizations to Diesel schema for org membership lookup
- Implement check_bot_access(): public bots allow all users, private bots
require user membership in the bot's organization via user_organizations
- Wire access check into websocket_handler before WS upgrade
- Read is_public from bots table instead of bot_configuration
- Add database_name field to Bot model
- Added SWITCHER_TOGGLE message type (8) for reprocessing last user message with active switchers
- Backend: Handler fetches last user question from DB, mutates message in-place, injects switcher prompts into system_prompt
- Backend: Switcher replays skip message_history save to avoid duplication
- Frontend: toggleSwitcher() sends SWITCHER_TOGGLE when input empty, sendMessage() when text present
- Frontend: Added TOOL_EXEC and SWITCHER_TOGGLE to MessageType constants
- Fixed session_id shadowing bug in DB query (used session_id_for_query)
- Preserves conversation history for LLM context when reprocessing with switchers
