emisar
Let AI agents investigate and operate infrastructure through declared actions, bounded by policy and host-side checks. Approvals when required, with an audit tr
What is emisar?
Let AI agents investigate and operate infrastructure through declared actions, bounded by policy and host-side checks. Approvals when required, with an audit trail.
Overview of emisar
What it is
emisar is an open-source tool that lets AI agents investigate and operate infrastructure through declared actions, bounded by policy and host-side checks. It provides an MCP (Model Context Protocol) surface for agents, replacing direct shell access with a catalog of approved actions. The system includes a control plane for policy evaluation and an outbound-only runner on each host that verifies and executes actions. It is built by Protectorate.
Key features
- Agents discover and request actions via MCP, with typed arguments and schema validation.
- Policy on the control plane decides which actions run, require human approval, or are denied.
- An outbound-only runner on each host verifies pack hashes, validates arguments, and enforces local limits before execution.
- Action packs define the executable, argument schema, risk level, timeout, output limits, redaction rules, and side-effect description.
- The runner opens only an outbound TLS WebSocket and exposes no inbound listener.
- Packs are content-addressed; the runner recomputes the hash from disk before execution.
- The audit trail includes denied and pending requests, and each runner writes a hash-chained JSONL journal of execution attempts.
Use cases
- Allowing AI agents to perform infrastructure investigation and operations with bounded authority.
- Enforcing policy-based approvals for high-risk actions on production systems.
- Providing an auditable, declarative interface for agent-driven system administration.
- Enabling operators to add new capabilities (packs) without deploying additional MCP servers or reconfiguring agents.
Getting started
See the repository README for the full walkthrough. The quick start for one host requires an emisar account, a Linux host with systemd and sudo, and outbound HTTPS access to specific domains. The installer command is:shcurl -fsSL https://emisar.dev/install.sh | sudo EMISAR_ENROLLMENT_KEY=emkey-enroll-... bash
License and pricing
The license is not stated in the provided metadata. The project is listed as open source.
這是什麼
emisar 是一個開放原始碼工具,讓 AI 代理程式透過宣告式動作來調查與操作基礎設施,並以政策與主機端檢查作為邊界。它為代理程式提供 MCP(模型上下文協定)介面,以核准的動作目錄取代直接 shell 存取。系統包含一個用於政策評估的控制平面,以及每個主機上僅對外連線的執行器,負責驗證與執行動作。由 Protectorate 開發。
主要功能
- 代理程式透過 MCP 發現並請求動作,支援型別參數與 schema 驗證。
- 控制平面上的政策決定哪些動作可執行、需要人工核准或遭到拒絕。
- 每個主機上的僅對外連線執行器在執行前會驗證套件雜湊、驗證參數並強制執行本機限制。
- 動作套件定義了可執行檔、參數 schema、風險等級、逾時、輸出限制、遮罩規則與副作用描述。
- 執行器僅建立對外的 TLS WebSocket,不開放任何監聽連線。
- 套件以內容定址;執行器在執行前會從磁碟重新計算雜湊。
- 稽核軌跡包含被拒絕與待處理的請求,每個執行器也會將執行嘗試寫入雜湊鏈結的 JSONL 日誌。
使用情境
- 讓 AI 代理程式在有限權限下執行基礎設施調查與操作。
- 對生產系統上的高風險動作強制執行基於政策的核准流程。
- 為代理程式驅動的系統管理提供可稽核的宣告式介面。
- 讓操作人員新增功能(套件)時無需部署額外的 MCP 伺服器或重新設定代理程式。
快速開始
完整操作說明請參閱儲存庫 README。單一主機的快速入門需要 emisar 帳號、具備 systemd 與 sudo 的 Linux 主機,以及對特定網域的 HTTPS 對外連線。安裝指令為:shcurl -fsSL https://emisar.dev/install.sh | sudo EMISAR_ENROLLMENT_KEY=emkey-enroll-... bash
授權與價格
提供的元資料中未說明授權條款。此專案標示為開放原始碼。
Based on the project's public README and repository metadata.