Plans deserve their own environment.
HeySpex is where plans get developed, challenged, and settled. Your IDE is where they get built. Keeping those two apart is not a limitation. It is the design.
The agent that builds should not be the agent that plans.
Plan modes inside IDEs feel convenient and fail quietly. The planning context is polluted by the implementation context. The plan gets rubber-stamped by the thing that wants to start typing. And nothing forces it to be settled before the build starts, so "plan" becomes a preamble instead of a gate.
Code solved this problem decades ago: drafts, reviews, versions, approvals. Specs deserve the same development discipline, in their own environment. That is what HeySpex is.
Built to sit inside your Claude Code workflow.
The vault is not a tab you switch to. Hooks, skills, slash commands, and MCP tools put HeySpex inside the session: capture a thought mid-build, preflight a task before an agent claims it, pull the next settled task with its context already packed.
And nothing is locked to one vendor. Claude Code is the deepest integration, not the only one: Cursor, Cline, Continue, and whatever ships next month are first-class over the same open protocol. Switching agents never means losing your project's memory.
The queue is plain code. That is the safety.
The core problem has not changed: AI agents are good at executing the wrong thing perfectly. HeySpex's answer is to keep the judgment layer deterministic. The scheduler, the gates, and the ranking are plain code over the markdown in your vault. Same vault, same answer, every time, with the reason named.
When "what should happen next, and is it safe to start" has a provable answer, handing dispatch to a nondeterministic agent stops being reckless. The agents stay creative. The ground they stand on does not move.
The plans are only half of it.
What the agents read, and why retrieval never misses, is covered in the second brain.