为什么值得关注

它拥有图原生的任务存储，支持依赖跟踪、原子声明和就绪任务检测，嵌入式模式下零基础设施，并可通过标准化 schema 对接 GitHub、Jira、Linear 等多个后端。

适合谁使用

AI 智能体开发者
构建智能体工作流的运维团队
对任务编排感兴趣的开源爱好者
使用 MCP 兼容 LLM 客户端的开发者

典型使用场景

管理 AI 编码智能体的复杂任务依赖
通过依赖图协调多智能体工作流
将 GitHub、Jira 或 Linear 的任务管理集成到智能体流水线
利用 MCP 工具让 LLM 认领和解决任务

项目优势

使用 SQLite 的零依赖嵌入式模式，单二进制文件
丰富的图查询（关键路径、瓶颈、依赖树）
提供 10 个工具的 MCP 适配器，方便 LLM 集成
美观的 Web 仪表盘，包含看板、有向无环图和检查器

使用前须知

处于早期实验阶段（README 中已注明）
本地开发需要手动安装 Node.js 和 Go
客户端-服务器模式仅支持 PostgreSQL（未提及其他数据库）
未提及身份验证或多租户支持

README 快速开始

🚜 Farm Table

Farm Table is an open-source task runtime built specifically for AI agents. It provides coding and execution agents with a single, predictable, and robust interface to receive work, understand complex dependency structures, and track execution progress—whether tasks reside in GitHub, Jira, Linear, or Farm Table's own built-in graph-native backend.

When no external platform is connected, the built-in backend provides a graph-native task store complete with dependency tracking, atomic CAS (Compare-And-Swap) claims, and instant ready-task detection. No API keys, no SaaS accounts, and zero complex configuration are required.

NOTE: Farmtable is at an early and experimental stage

🚀 System Architecture & Modes

Farm Table operates transparently in two modes from the exact same codebase:

┌─ Embedded Mode (Default) ──────────────────────────────────────┐
│  ft CLI ──bufconn──► In-Process FarmTableService               │
│                      └──► EntStore (SQLite)                    │
│  Single process. Zero infrastructure. Just works.              │
├─ Client-Server Mode ───────────────────────────────────────────┤
│  ft CLI ──gRPC/TLS──► farmtable-server (Separate Process)      │
│                       └──► EntStore (Postgres)                 │
│  Multi-agent coordination. Production deployments.             │
└────────────────────────────────────────────────────────────────┘

Embedded Mode: Automatically selected by default. SQLite is used as the local store at ~/.farmtable/farmtable.db (overridable via FARMTABLE_DB_PATH). Communication is routed through bufconn (an in-memory network connection), creating a single-binary zero-dependency experience.
Client-Server Mode: Enabled automatically if the FARMTABLE_SERVER environment variable, --server flag, or the config file server key is supplied. This connects the CLI or agent to a standalone farmtable-server backed by Postgres for multi-agent coordination.

💎 Core Abstractions & Features

📦 Normalized Task Object (NTO)

Every task—whether synced from GitHub or managed locally—is represented in a single, predictable schema defined in Protobuf (proto/farmtable/v1/farmtable.proto).

Task State Model: Native tasks use asserted stages such as triage, accepted, working, h

farmtable