# Seshat > Open-source Go agent runtime for autonomous AI systems and multi-agent automation. MCP, RAG, 15+ LLM providers and 60+ built-in tools in a single binary. - **Type:** MCP server - **Install:** `agentstack add mcp-kpo-tech-seshat` - **Verified:** Pending review - **Seller:** [KPO-Tech](https://agentstack.voostack.com/s/kpo-tech) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** Apache-2.0 - **Upstream author:** [KPO-Tech](https://github.com/KPO-Tech) - **Source:** https://github.com/KPO-Tech/seshat - **Website:** https://seshat-ai.com/docs ## Install ```sh agentstack add mcp-kpo-tech-seshat ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About Seshat Open-source Go Agent Runtime for autonomous AI systems Built for the age of agent teams, not just single agents. 🌐 Website Β· πŸ’¬ Discussions Β· πŸ› Issues Β· πŸ–₯️ seshat-ai --- ## πŸ€” Why Seshat The agentic AI space is moving fast. Dozens of frameworks have emerged β€” LangChain, LangGraph, CrewAI, AutoGen, and many others. Almost all of them are Python-first. That works fine for experimentation, but it creates real friction the moment you want to ship something: dependency hell, slow startup, heavy containers, difficult embedding in existing services. **⚑ Go changes the equation.** A single statically linked binary. No runtime, no venv, no dependency manager. Low latency, predictable memory, easy deployment anywhere β€” from a Raspberry Pi to a Kubernetes cluster. Seshat brings that to agentic AI. But the choice of language is the smaller part of the story. --- ### 🚧 The real gap: agents don't know how to work together Individual agents have become impressive. They can reason for hours, use tools, write code, search the web, produce deliverables. The frontier has moved. The hard unsolved problem is no longer *what a single agent can do* β€” it is **how multiple agents can work together in a coherent, observable, and economically useful way**. Most current systems handle this with pipelines, handoffs, or orchestration graphs. These are useful, but they remain technical constructs. They don't model *work* the way humans actually do it. --- ### 🧠 The thesis behind this project We believe the right model for agent collaboration is not the workflow graph β€” it is the **human organization**. Real teams don't communicate through function calls. They have roles, responsibilities, inboxes, meetings, status reports, shared decisions, and a working memory of the mission. A project lead delegates tasks. A specialist delivers a report. A reviewer pushes back. A coordinator synthesizes. Information flows explicitly. Decisions are recorded. Nothing is lost in a black-box state machine. Seshat is being built around this idea: **study how humans work in teams β€” in companies, research groups, or any collaborative setting β€” and reproduce those coordination patterns with autonomous agents.** This means agents that: - πŸ“¬ communicate through structured messages, not just function calls - πŸ“‹ can send a report, schedule a review, claim a task, or escalate a blocker - 🧩 have an identity, a role, and a history β€” not just a prompt and a response - πŸ—‚οΈ share a mission memory that keeps decisions, constraints, and objectives alive across the team --- ### πŸ—ΊοΈ Where we are and where we're going We are building the foundation. Seshat today is a solid, production-grade runtime: persistent sessions, 60+ tools, 15 LLM providers, MCP client, RAG, skills, gRPC API, sandboxed execution. It is already useful as a standalone CLI, an embeddable Go SDK, or the engine behind [seshat-ai](https://github.com/EngineerProjects/seshat-ai). The team primitives (mailboxes, roles, missions, structured inter-agent communication) are the next major layer. We are designing them now, informed by how real organizations function. Beyond these layers, the shape of the platform will be defined as much by the community as by any central roadmap. Domain experts in healthcare, education, legal, finance, logistics, or any other professional field will propose and contribute specialized integrations tailored to the realities of their industry. Seshat is not a product with a fixed scope. It is an OS for agentic services: an open foundation on which any application can be built for any use case, by the people who understand that use case well enough to build it right. --- ### 🀝 An open project: all ideas welcome **This is an open research and engineering project as much as a software one.** The ecosystem we want to build is ambitious, and no single perspective has all the answers. If you have ideas about how agents should collaborate, how human team structures map to agentic systems, what organizational patterns are worth modeling, or what the ecosystem is still missing: your input is genuinely valuable. πŸ‘‰ Open a [Discussion](https://github.com/EngineerProjects/seshat/discussions), an issue, or a PR. Every suggestion is read. We are building this together. πŸš€ --- ## πŸ–₯️ Terminal UI `seshat chat` drops you into a full-featured terminal interface built for long-running agent sessions. Welcome β€” up and running in 10 seconds Settings panel β€” every shortcut, one keystroke away (ctrl+p) Model selection β€” switch across 20+ models in 2 keystrokes Provider config β€” API keys encrypted, scoped per provider Agent at work β€” full reasoning trace, thinking blocks + tool calls Streaming results β€” streamed responses with markdown + tool timings > **Keyboard shortcuts:** `ctrl+p` settings Β· `ctrl+m` models Β· `ctrl+s` sessions Β· `ctrl+,` providers Β· `ctrl+n` new session Β· `ctrl+t` tasks Β· `ctrl+u` copy last response Β· `ctrl+y` toggle yolo Β· `ctrl+o` open editor Β· `ctrl+g` help Β· `ctrl+c` quit > **Clipboard note (Linux):** selection copy works best when `wl-clipboard` (Wayland) or `xclip`/`xsel` (X11) is installed. Without a system clipboard backend, Seshat can request terminal clipboard access but cannot guarantee a real system copy. > **Compaction note:** transcript compaction is automatic today. A dedicated manual compact action is planned for the TUI once the runtime exposes a real manual-compaction hook. --- ## 🌐 The Seshat Ecosystem seshat is the **headless runtime**: pure Go, no UI, no users, no billing. It is the foundation everything else builds on. ### πŸ–₯️ seshat-ai β€” Desktop & Platform **[β†’ seshat-ai](https://github.com/EngineerProjects/seshat-ai)** is the full production platform built on top of this engine. If you want a ready-to-use application rather than a library, that is where you want to go. | | seshat (this repo) | seshat-ai | |---|---|---| | **What it is** | Go runtime + SDK + CLI | Desktop app + REST API platform | | **Stack** | Go | Go (API) + TypeScript/React/Electron (desktop) | | **License** | Apache 2.0 | AGPL-3.0 | | **Who it's for** | Developers embedding agents in their own apps | End users, teams, self-hosters | | **Includes** | Engine, tools, providers, gRPC, CLI/TUI | Multi-user auth, workspaces, knowledge base, scheduler, desktop UI | **What seshat-ai gives you today:** - πŸ–₯️ Native desktop app (Electron + React) with chat, tool views, plans, settings and a visual skills creator - πŸ‘₯ Multi-user backend with organizations, workspaces, per-user API keys, quotas and audit log - πŸ“‘ REST + SSE HTTP API compatible with the Anthropic `/v1/messages` format - πŸ“š Knowledge base with hybrid BM25 + vector search and file ingestion - ⏰ Scheduled tasks, memories, plans and MCP server management **Coming next:** - 🀝 Agent teams: persistent groups of specialized agents collaborating on shared missions, each with its own inbox, role and memory - πŸ€– Automation and background workflows triggered by schedule, events or voice - πŸ–ΌοΈ Image generation integrated directly into the chat and workspace - πŸŽ™οΈ Voice input and audio output so you can talk to your agents naturally - 🌐 A multi-workspace environment covering code, research, creation and learning, all sharing the same runtime and data layer ### 🀝 Contribution split | If you want to... | Contribute to... | |---|---| | Improve execution speed, reduce latency, optimize the agent loop | **seshat** (Go) | | Add a new LLM provider or tool | **seshat** (Go) | | Expose new capabilities in the SDK or gRPC API | **seshat** (Go) | | Improve the desktop UI, add new views, fix UX | **seshat-ai** (TypeScript/React) | | Build features like agent teams, automation or scheduling | **seshat-ai** (Go API + React) | The engine is intentionally kept minimal and fast. If you need something from the SDK that is not exposed yet, open an issue and we will prioritize it. ### πŸ“¦ Installation **End users β€” one command, fully configured:** ```bash curl -fsSL https://raw.githubusercontent.com/EngineerProjects/seshat/main/scripts/install.sh | bash ``` Downloads the right binary for your platform, adds it to your PATH, installs `uv` and `docling-serve` for document processing, and leaves the runtime directory (`~/.config/seshat-cli/`) ready. The DB and sessions are created on first run. Options: ```bash NO_PYTHON=1 bash # resume a specific session seshat chat --continue # resume the most recent session seshat run "list all TODO comments in this codebase" # one-shot task seshat sessions list # browse past sessions seshat setup --check # show uv / docling status seshat version # print installed version seshat help # full command reference ``` Sessions are persisted locally in SQLite. Skills are loaded from `.seshat/skills/` in your project. The full tool set is available: file edits, sandboxed bash, web search, browser, MCP servers, sub-agents. --- ### 2. 🌐 gRPC Server Run seshat as a gRPC service and generate clients for any language. ```bash # Development ANTHROPIC_API_KEY=sk-ant-... go run ./cmd/grpc # From build ANTHROPIC_API_KEY=sk-ant-... ./bin/seshat-grpc ``` Server starts on `:50051`. The contract lives in `pkg/grpc/proto/nexus.proto`. Generate a client for Python, TypeScript, Java, Rust, or any gRPC-supported language: ```bash # Python python -m grpc_tools.protoc -I pkg/grpc/proto --python_out=. --grpc_python_out=. nexus.proto # TypeScript npx grpc-tools --js_out=. --grpc_out=. pkg/grpc/proto/nexus.proto ``` One runtime. Every language. --- ### 3. πŸ“¦ Go SDK Embed the full runtime in your own Go application. ```bash go get github.com/EngineerProjects/seshat/pkg/sdk ``` ```go import "github.com/EngineerProjects/seshat/pkg/sdk" client, err := sdk.NewClient(&sdk.ClientConfig{ APIKey: os.Getenv("ANTHROPIC_API_KEY"), Model: sdk.ModelIdentifier{Provider: "anthropic", Model: "claude-sonnet-4-20250514"}, }) if err != nil { log.Fatal(err) } defer client.Close() session, _ := client.CreateSession(ctx) resp, _ := session.SubmitMessage(ctx, "Write a Go HTTP handler for /health") fmt.Println(resp.Content) ``` --- ## πŸ“Š How Seshat Compares | Feature | **seshat** | Claude Agent SDK | OpenAI Agents SDK | LangGraph | CrewAI | |---|:---:|:---:|:---:|:---:|:---:| | Language | **Go** | Python/TS | Python | Python | Python | | Single binary (no deps) | βœ… | ❌ | ❌ | ❌ | ❌ | | CLI included | βœ… | ❌ | ❌ | ❌ | ❌ | | gRPC server (any language) | βœ… | ❌ | ❌ | ❌ | ❌ | | Multi-provider | βœ… (15) | ❌ Claude only | ❌ OpenAI only | βœ… | βœ… | | MCP client | βœ… | βœ… | βœ… | Partial | ❌ | | Sandboxed bash (Landlock) | βœ… | βœ… | ❌ | ❌ | ❌ | | Skills system | βœ… | ❌ | ❌ | ❌ | ❌ | | Built-in RAG | βœ… | ❌ | ❌ | ❌ | ❌ | | Browser automation | βœ… | βœ… | ❌ | ❌ | ❌ | | Session persistence | βœ… | ❌ | ❌ | βœ… | ❌ | | OTel tracing | βœ… | ❌ | ❌ | βœ… | ❌ | | Open-source license | Apache 2.0 | MIT | MIT | MIT | Apache 2.0 | --- ## ✨ Capabilities | Capability | Details | |---|---| | 🌍 **Multi-provider** | 15 providers: Anthropic, OpenAI, Gemini, Mistral, DeepSeek, Ollama, OpenRouter, AWS Bedrock, GCP Vertex, Azure Foundry, Codex, MiniMax, Z.ai, OpenCode, Cloudflare Workers AI | | πŸ› οΈ **60+ built-in tools** | File read/write/patch, bash (Landlock sandbox), web search, web fetch, browser (Playwright), grep/glob, LSP, sub-agents, RAG, tasks, memory, worktree, notebooks, image generation, TTS/STT | | πŸ”Œ **MCP client** | Universal MCP client: plug in any MCP server (GitHub, Postgres, Slack, Docker, Notion, ...) | | ⚑ **Skills** | Markdown instruction files injected into the system prompt: encode your team's conventions and domain expertise | | 🎯 **Execution modes** | `execute` (default), `plan` (review before act), `pair_programming` (collaborative) | | πŸ”’ **Permission engine** | Per-tool deny rules, auto-mode LLM classifier, configurable per session (`auto` / `acceptEdits` / `onRequest` / `bypass` / `never`) | | πŸ’Ύ **Session persistence** | SQLite-backed multi-turn sessions, resumable across restarts | | πŸ“‘ **Streaming** | Text chunks + structured runtime events (tool calls, plan events, permission requests, token usage) | | 🧠 **Long-context compaction** | Automatic context compression when approaching the model's window (configurable threshold) | | πŸ“‰ **Observability** | Prometheus metrics + OpenTelemetry tracing (OTLP gRPC export, no-op when endpoint not set) | --- ## πŸ—‚οΈ Repository Structure ``` seshat/ β”œβ”€β”€ cmd/ β”‚ β”œβ”€β”€ cli/ ← seshat CLI entrypoint (TUI + one-shot commands) β”‚ └── grpc/ ← gRPC server entrypoint β”œβ”€β”€ pkg/ ← public API (safe to import from outside) β”‚ β”œβ”€β”€ sdk/ ← Go SDK: Client, sessions, streaming, callbacks β”‚ β”œβ”€β”€ types/ ← shared types: Message, ToolUse, TokenUsage, ... β”‚ β”œβ”€β”€ agent/ ← agent definitions, built-in registry β”‚ β”œβ”€β”€ providers/ ← LLM provider abstraction, routing, fallback β”‚ β”œβ”€β”€ mcp/ ← MCP client: stdio, SSE, HTTP transports β”‚ β”œβ”€β”€ rag/ ← chunking, embedding, hybrid vector search β”‚ β”œβ”€β”€ skills/ ← skill loading, frontmatter parsing, injection β”‚ β”œβ”€β”€ memory/ ← in-session state, compaction strategies β”‚ β”œβ”€β”€ web/ ← web search, fetch, browser (Playwright) β”‚ β”œβ”€β”€ storage/ ← artifact store: S3, local filesystem β”‚ β”œβ”€β”€ vector/ ← vector DB abstraction β”‚ β”œβ”€β”€ contract/ ← Tool interface, CallResult, registry β”‚ β”œβ”€β”€ auth/ ← provider auth abstraction, OAuth device flow β”‚ β”œβ”€β”€ workspace/ ← sandbox path resolution, workspace layout β”‚ β”œβ”€β”€ monitoring/ ← Prometheus metrics, OTel spans β”‚ β”œβ”€β”€ docling/ ← PDF/DOCX/audio conversion via docling-serve β”‚ β”œβ”€β”€ grpc/ ← proto definitions and generated code β”‚ └── config/ ← app-level config from env └── internal/ ← private implementation (do not import directly) ``` > seshat-ai and any third-party consumer must import `pkg/*` only, never `internal/*`. --- ## 🌐 Supported Providers | Provider ID | Service | Auth | |---|---|---| | `anthropic` | Anthropic | `ANTHROPIC_API_KEY` | | `openai` | OpenAI | `OPENAI_API_KEY` | | `gemini` | Google Gemini | `GOOGLE_API_KEY` | | `mistral` | Mistral AI | `MISTRAL_API_KEY` | | `deepseek` | DeepSeek | `DEEPSEEK_API_KEY` | | `ollama` | Ollama (local) | none | | `openrouter` | OpenRouter | `OPENROUTER_API_KEY` | | `bedrock` | AWS Bedrock | `AWS_ACCESS_KEY_ID` + region | | `vertex` | GCP Vertex AI | `ANTHROPIC_VERTEX_PROJECT_ID` + region | | `foundry` | Azure AI Foundry | `ANTHROPIC_FOUNDRY_API_KEY` | | `codex` | ChatGPT Pro (OAuth) | device-code flow | | `minimax` | MiniMax | `MINIMAX_API_KEY` | | `z-ai` | Z.ai | `Z_AI_API_KEY` | | `opencode` | OpenCode Zen | `OPENCODE_API_KEY` | | `workers-ai` | Cloudflare Workers AI | `CLOUDFLARE_API_KEY` | Full model listings and capabilities: [`docs/providers.md`](./docs/providers.md). --- ## πŸš€ Quick Start ```bash # 1. Install curl -fsSL https://raw.githubusercontent.com/EngineerProjects/seshat/main/scripts/install.sh | bash # Reload your shell (or open a new terminal) if prompted, then: # 2. Configure your provider seshat config --provider anthropic --api-key sk-ant-... seshat config --model anthropic:claude-sonnet-4-20250514 # 3. Start chatting seshat chat # new session (opens TUI) seshat chat --continue # resume last session seshat chat --resume # resume a specific session seshat run "list all TODO comments in this codebase" # one-shot task ``` > **No API key?** Use Ollama for free local infere … ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source β€” we do not rehost the code. - **Author:** [KPO-Tech](https://github.com/KPO-Tech) - **Source:** [KPO-Tech/seshat](https://github.com/KPO-Tech/seshat) - **License:** Apache-2.0 - **Homepage:** https://seshat-ai.com/docs Install and usage instructions live in the source repository linked above. ## Pricing - **Free** β€” Free ## Security capabilities Automated source analysis of v0.1.0 β€” what this tool can access: - **Network access:** yes - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** yes - **Dynamic code execution:** no *"Yes" means the capability is present in the source β€” more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** β€” security scan: flagged β€” Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-kpo-tech-seshat - Seller: https://agentstack.voostack.com/s/kpo-tech - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack β€” the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.