Vibe-Kanban-Alternative – Classic AI agent Kanban with mem0 persistent memory

forked from dexloom/vibe-kanban-indie Notifications You must be signed in to change notification settings Fork 0 Star 1 BranchesTags Open more actions menu Latest commit History 2,456 Commits 2,456 Commits Folders and files NameName Last commit message Last commit date .cargo .cargo .github .github apps/macos apps/macos assets assets automation automation crates crates dev_assets_seed dev_assets_seed docs docs npx-cli npx-cli packages packages patches patches scripts scripts shared shared .dockerignore .dockerignore .env.example .env.example .gitignore .gitignore .npmrc .npmrc AGENTS.md AGENTS.md CHANGELOG.md CHANGELOG.md CLAUDE.md CLAUDE.md CODE-OF-CONDUCT.md CODE-OF-CONDUCT.md CONTEXT.md CONTEXT.md CONTRIBUTORS.md CONTRIBUTORS.md Caddyfile.example Caddyfile.example Cargo.lock Cargo.lock Cargo.toml Cargo.toml Dockerfile Dockerfile LICENSE LICENSE Makefile Makefile PUBLISHING.md PUBLISHING.md README.md README.md SPEC-conversation-list.md SPEC-conversation-list.md SPEC.md SPEC.md TODO.md TODO.md gitea.toml.example gitea.toml.example local-build.sh local-build.sh mobile-testing.md mobile-testing.md package.json package.json pnpm-lock.yaml pnpm-lock.yaml pnpm-workspace.yaml pnpm-workspace.yaml projects.toml.example projects.toml.example restart.sh restart.sh rust-toolchain.toml rust-toolchain.toml rustfmt.toml rustfmt.toml Repository files navigation The resilient, 100% self-hosted cockpit for single-developer AI orchestration. Persistent Graph Memory (mem0) • 10+ Coding Agents (inc. Antigravity/AGY) • Gitea/Forgejo PRs • TUI Cockpit • Telegram Daemon • Usage Observability Quick Start • Overview • What's Different • Project Memory (mem0) • Supported Agents • Chat & Controls • Usage Dashboard • TUI Cockpit • Telegram Bridge • Development ⚡ Quick Start (Instant Run) Launch the entire cockpit with a single command — no install, no accounts, no cloud setup: npx vibe-kanban-alternative 💡 Zero Setup Required: Downloads prebuilt binaries and launches the local web cockpit directly at http://localhost:3001 (with backend on :3002). ⚙️ Full Setup & Environment Configuration For development, custom ports, local mem0 vector stores, and custom agent configurations: 1. Clone & Install git clone https://github.com/flashlan/vibe-kanban-alternative.git cd vibe-kanban-alternative pnpm i 2. Configure Environment Variables Duplicate the template configuration file: cp .env.example .env Open .env in your editor and configure your environment: # ========================================== # 🌐 CORE SERVER CONFIGURATION # ========================================== PORT=3000 HOST=localhost NODE_ENV=development # ========================================== # 🧠 MEM0 LONG-TERM MEMORY SETTINGS # ========================================== MEM0_ENABLED=true MEM0_API_KEY=your_mem0_api_key_here # If running local embeddings/vector store: # MEM0_VECTOR_STORE=qdrant # MEM0_HOST=http://localhost:6333 # ========================================== # 🤖 AGENT API KEYS & RUNTIMES # ========================================== ANTHROPIC_API_KEY=sk-ant-... OPENAI_API_KEY=sk-... GEMINI_API_KEY=... # Antigravity (AGY) Specific Settings AGY_CLI_PATH=/usr/local/bin/agy AGY_DEFAULT_TEMPERATURE=0.2 # ========================================== # 📊 METRICS & TELEMETRY # ========================================== ENABLE_METRICS_DASHBOARD=true METRICS_STORAGE_PATH=./data/metrics.sqlite # ========================================== # 💾 BACKUP CONFIGURATION # ========================================== BACKUP_ENABLED=true BACKUP_INTERVAL_MINUTES=30 BACKUP_STORAGE_PATH=./backups 3. Start Development Cockpit ./restart.sh 📌 Context: Reviving & Evolving Vibe-Kanban Following the official cloud shutdown of Bloop's hosted servers, developers were left with orphaned workspaces and broken dependencies. vibe-kanban-alternative is an actively maintained, independent evolution of BloopAI/vibe-kanban and dexloom/vibe-kanban-indie. Built specifically for a single-developer workflow, it requires no cloud accounts, no team auth, and zero remote telemetry. You run it entirely on your machine, orchestrating a fleet of coding agents via browser, terminal (TUI), or phone (Telegram). ⚡ What's Different in This Fork? Capability Upstream (Bloop) Indie Baseline ⚡ vibe-kanban-alternative (This Fork) Server Infrastructure 🛑 Sunsetting / Dead 🟢 Local / Self-hosted 🟢 100% Offline & Local Runtime Cross-Session Memory ❌ Ephemeral (Lost per card) ❌ None 🧠 Native mem0 + Qdrant + NetworkX Graph Memory Prompt Cache-Hit Architecture ❌ None ❌ None ⚡ Deterministic, Cache-Friendly Memory Prefix Injection Telemetry & Observability ❌ None ⚠️ Minimal 📊 Full Settings → Usage Dashboard (Tokens, Heatmaps, Agents) Coding Agents Matrix Legacy CLI subset Standard Set 🚀 10+ Agents: Claude Code, Antigravity (AGY), Codex, Gemini CLI, etc. Antigravity (AGY) Agent ❌ None ⚠️ Basic text mode 💎 Full stream-json parsing, ToolUse visual cards & reasoning effort control Chat Input & History ⚠️ Basic textarea Standard WYSIWYG ⌨️ Terminal-style ArrowUp/Down prompt history + Enter/Ctrl+Enter modes Self-Hosted Git Remotes GitHub Only GitHub + Basic Gitea 🐙 Auto-routing Gitea / Forgejo REST API + GitHub (gh) Remote Control & Cockpit Web UI only TUI + Telegram 📱 Terminal TUI Cockpit + Send-Only Telegram Forum Bridge Backup & Disaster Recovery ❌ None Basic 💾 Full DB, Settings & mem0 Vector/Graph State Export/Import Chat & UI Streaming ⚠️ Latency bugs on long diffs Standard ⚡ Optimized Canvas/Chat rendering & worktree panel fixes 🚀 Overview Software engineering increasingly means directing coding agents — planning work, spawning a model to implement it, reviewing its diff, and shipping. The bottleneck is no longer typing code; it's orchestrating, reviewing, and keeping many agent sessions coherent. vibe-kanban-alternative is built to make that process fast, local, and personal: a single developer, entirely on their own machine, no team, no cloud, no account. At its core it's a kanban board that plans and tracks agent work, plus a workspace runtime that turns each card into a real branch, terminal, and dev-server where any of 10+ coding agents (Claude Code, OpenCode, Qwen Code, Codex, Gemini CLI, Antigravity, Copilot, Amp, Cursor, Droid, CCR) executes the plan: Plan with kanban issues — boards, columns, priorities, tags, sub-issues, pipelines; cards are the single source of truth for a piece of work. Run coding agents in workspaces — each card launches a workspace: a branch, a terminal, a dev server, and an agent following a configurable pipeline (Quick, Basic, async variants). Review diffs and iterate — inline comments, diffs, preview browser, and the manual-review stage that pauses the agent and raises an alarm so you approve the result before any merge or PR. Switch between 10+ coding agents — drive Claude Code, OpenCode, Qwen Code, Codex, Gemini CLI, Antigravity (agy), Copilot, Amp, Cursor, Droid, and CCR from one board. Cross-session project memory (mem0) — agents recall and persist verified facts about the repositories they work in, keyed per repository and shared across CLIs, with a graph memory that survives restarts. Usage & observability — a Settings → Usage dashboard with per-day activity, per-agent execution bars, extraction-token monitoring, and project progress. Workspaces, PRs, and merge — dispatch work to existing sessions, open PRs (GitHub or Gitea/Forgejo) with AI-generated descriptions, squash-merge to base. Terminal & phone — a TUI cockpit and Telegram escalation keep you in control without the browser. 🧠 Project Memory (mem0) Cross-session memory for every coding agent. vibe-kanban-alternative ships with a first-class mem0 integration so the agents that drive your workspaces share a durable, semantic memory of the repositories they work in. Core Capabilities: ⚬ Automatic Recall on Launch: Workspaces query stored memory for the current repository slug and prepend it to the agent prompt. Hard-won architectural conventions and debugging discoveries are never forgotten. ⚬ Shared Repository Knowledge: Memory is keyed per repository. You can start a task on Claude Code and switch to OpenCode or Antigravity mid-project without losing context. ⚬ Verified Fact Save-Back: The memory pipeline stage instructs the agent to persist only self-contained, verified facts (architectural decisions, patterns, root causes) via memory_save. Ephemeral chatter is filtered out. ⚬ MCP Tools Integration: Agents have access to memory_search, memory_recall, and memory_save as first-class Model Context Protocol (MCP) tools. ⚬ Graph-Based Memory (GraphML): Entity and relation extraction creates an interconnected graph (mem0 + Qdrant + NetworkX), persisted on disk (/data/graphs/*.graphml) so knowledge structures survive container reboots. ⚡ Prompt Cache-Hit Design To minimize token costs on providers with prompt caching (Anthropic, OpenRouter, DeepSeek): Static Prefix Placement: The memory block is injected into the static task prefix before dynamic user instructions. Deterministic Ordering: Memory entries are sorted deterministically, ensuring byte-identical prompt prefixes across sessions for continuous cache hits. No Mid-Session Perturbation: memory_search and memory_save execute as MCP tool calls rather than system prompt mutations, keeping the token cache warm. mem0 Setup The project memory layer runs on a local Docker stack (mem0-vk): a mem0 API server on :8000, a Qdrant vector store, and a Python embeddings + NetworkX graph service. cd mem0-vk cp .env.example .env # then set an extraction LLM key (see below) docker compose up -d --build Configure from the app: Open Settings → Memory to manage the graph at runtime, configure extraction providers (Groq, OpenRouter, Local llama), and view token usage. 🤖 Supported Coding Agents vibe-kanban-alternative integrates natively with 10+ leading coding agents: Google Antigravity (agy) ⚡ (New): Full stream-JSON protocol support. Native visual cards for file inspection (view_file), search (grep_search, find_by_name), bash commands (run_command), and file edits (write_to_file, replace_file_content). Reasoning effort controls (Low, Medium, High) with automatic fallback for gemini-3.7-flash. YOLO mode auto-permission bypass (--dangerously-skip-permissions). Anthropic Claude Code: Headed & headless modes, full MCP tool approvals, and turn navigation. OpenCode & OpenCode Headed: Multi-model agent runner with local & remote inference. OpenAI Codex: Deep reasoning & plan generation. Qwen Code: High-performance local and cloud agent workflows. Google Gemini CLI: Native Gemini execution. GitHub Copilot CLI, Cursor Agent, Droid, and Amp. ⌨️ Chat & Terminal Interaction Prompt History Navigation (ArrowUp / ArrowDown): Press ArrowUp at the start of the chat box to cycle backward through previously sent commands and prompts. Press ArrowDown to cycle forward and seamlessly restore your uncommitted draft text. History is persisted locally across browser sessions. Customizable Send Shortcuts: Enter Mode: Press Enter to send instantly; use Ctrl + Enter, Cmd + Enter, or Shift + Enter to insert a newline. ModifierEnter Mode: Press Cmd/Ctrl + Enter to send; Enter for newline. Configurable under Settings → General. 📊 Usage & Observability Dashboard A per-machine activity dashboard in Settings → Usage shows how your agent time is spent, straight from the local database — no cloud telemetry: Totals — executions, agent time, and open issues over the last 30 days. GitHub-style act [truncated for AI cost control]