Blog
Notes from the alineo team.
What we ship in the sandbox substrate, the agent SDK, and the docs themselves.RSS
How to stop a runaway AI agent swarm
Pause, steer, or stop a whole tree of sandboxed AI agents from outside them. Real calls, real timings, and what these controls can't reach.
ProductAn MCP server for swarm control
alineo-mcp puts alineod's swarm-control API behind stdio, so any MCP client — Claude Code, Claude Desktop, Cursor — can spawn, steer, and pause sandboxed agent swarms as tool calls.
ProductA control plane for swarms of agents
alineod spawns, steers, and recovers trees of sandboxed agents over HTTP and SSE, so orchestrating a swarm isn't code you write yourself.
DocsThe docs are now built for agents
Every alineo docs page is available as clean Markdown, indexed in llms.txt, and one click away from ChatGPT, Claude, or Cursor.
EngineeringKeeping secrets out of agent sandboxes
An agent that runs code and reaches the network is one prompt injection from leaking its credentials. alineo injects them at the network edge, so the sandbox never holds the value.
ProductA permission gate for sandboxed agents
AgentSpec.permissions puts a human in front of an agent's tool calls — a one-word mode or a full per-tool policy, enforced on every call before it runs.