Skip to content

Project Introduction

CollabVibe is a collaborative programming orchestration engine that connects IM platforms with AI agent backends. Its core capabilities include:

  • IM message and interactive-card integration
  • Multi-backend agent execution
  • approval-driven human-in-the-loop workflows
  • local persistence for threads, snapshots, audit data, and state

Design goals

ThemeDescription
Human-in-the-LoopHigh-risk actions enter an approval flow and continue only after a user decision
Collaborative developmentContinuous execution, review, merge, and snapshotting around a thread
Local data retentionSQLite, logs, configuration, and workspace state are stored locally

Platform support

PlatformStatusCurrent capabilityCode location
Feishu / LarkSupportedWebSocket events, messages, cards, bot menu, group and DM entry pointssrc/feishu/*, src/feishu/channel/*
SlackTODOOutput adapter and socket handler exist, but the application-layer main path is not fully wiredsrc/slack/*, src/slack/channel/*
MS TeamsTODOReserved as an extension direction; not integrated in the current repository

Backend support

The current codebase only supports backend access through API-driven transports.

BackendTransportModeStatusNotes
codexcodexAPISupportedConnected through the Codex protocol / stdio
opencodeacpAPISupportedConnected through ACP
claude-codeacpAPISupportedConnected through ACP
gemini-cliTBDPlannedNot integrated in the current code
trae-cliTBDPlannedNot integrated in the current code
bash
# Preview the documentation locally
npm run docs:dev

Authentication and authorization

Platform integration credentials

ItemDescription
FEISHU_APP_IDFeishu app ID
FEISHU_APP_SECRETFeishu app secret
FEISHU_SIGNING_SECRETFeishu event-signature secret
FEISHU_ENCRYPT_KEYFeishu encrypted-event configuration

In-system access control

ComponentPurpose
SYS_ADMIN_USER_IDSInitial import of system administrators
users tablePersistent source of system-level roles
RoleResolverResolves roles
authorize / command-guardCommand-level authorization checks

How it is used

StepDescription
1A user sends a message in IM or clicks a card
2The Platform layer parses the event and forwards it into unified intent dispatch
3The shared layer decides whether the request takes the platform-command or agent-command path
4The orchestrator resolves the thread, backend, and runtime config
5The backend executes and pushes intermediate state back through streaming events
6High-risk actions enter an approval flow
7Results, thread state, and audit data are written to local storage

Quick entry points

If this is your first time reading the project, start with the platform integrations and then move into the architecture section:

Data retained locally

CategoryDefault location
Main SQLite databasecollabvibe.db
Backend configurationconfig
Logslogs
Workspace / worktree / snapshotslocal code directory and derived worktrees
bash
ls -lah .
ls -lah logs