Private beta · invite only

AI builds what you mean, not what it guesses

HeySpex is the planning layer between your ideas and your coding agents. Messy thoughts become settled, sized plans grounded in your repo. Agents get the full context, and every branch comes back with a receipt.....

Invites go out in small batches while HeySpex is in private beta.

what's next⌘K
EPICHS-CKOT-EPICCheckout: billing onboarding flow
next up · 02unclaimed · ready to dispatch
Connect Stripe setup to onboarding state machine
HS-STRP-TASK · open · sized to one session
Open in docs
01HS-ONBS-TASKAdd onboarding state machineDone · moved on
02HS-STRP-TASKConnect Stripe setup to onboarding state machineNext up
03HS-WBHK-TASKHandle Stripe webhook retries with idempotency keysQueued
The problem

AI agents are good at executing the wrong thing perfectly.

Speed without alignment ships a flawless implementation of a misunderstanding. The faster AI helps you build, the harder the basic questions get:

01
Context scatters.
Across chats, commits, branches, notes, prompts, and your memory. What was the original intent? Why was this decision made?
02
Intent decays.
Each session starts from zero. The AI guesses, you correct, and the corrections evaporate. Next session, same guesses.
03
Trust erodes.
Without evidence, you have no idea whether AI-generated code actually matches what you asked for.
How it works

From messy idea to a branch that explains itself.

01Capture
Never lose the thought.

Say it, clip it, or type it the moment it occurs. Voice from the desktop companion, pages from the browser extension, notes from the console, or straight from inside an agent session. Everything lands in one pipeline.

voice · browser · console · in-session
02Plan
Settle the plan before anything builds.

Ideas become milestones, epics, and tasks, each sized to fit one agent session. Before work starts, HeySpex checks the plan against what your vault already knows, so new work never contradicts old decisions.

milestone → epic → task
03Dispatch
Always know what to work on next.

A deterministic scheduler picks the next task and says why it won. Hand it to Claude Code, Cursor, or any agent over MCP with the context already packed. Parallel agents coordinate through claims instead of stepping on each other.

▸ launch agent · claim held
04Verify
Every branch explains itself.

When the work lands, HeySpex shows plan versus actual: what was asked, what was built, and where they differ. Trust comes from the receipt, not from rereading the diff at midnight.

planned 7 · built 7 · drift 0
See the full pipeline, gates and all →
Second brain

A second brain for Claude Code.

Every plan, decision, and lesson lives in plain markdown with YAML frontmatter, in your repo. Open it in Obsidian and it just works. Ask for it over MCP and retrieval is lexical, fast, and exact, because HeySpex wrote the files it is searching. And links never rot: superseded specs point to their canonical successor.

markdown · frontmatter · [[wikilinks]] · mcp
Explore the second brain →
Agent workflows

Plans deserve their own environment.

Plan modes inside IDEs let the agent that wants to start typing grade its own plan. HeySpex develops specs separately from implementation, gates the build on a settled plan, and hands your coding agent the work through hooks, skills, slash commands, and MCP. There is no AI inside HeySpex: same vault, same answer, every time.

claude code · cursor · any mcp agent
See agent workflows →
Private beta

Invite only, on purpose.

HeySpex is being built with a small group of builders who use AI agents on real work every day. Small batches keep onboarding personal and feedback honest. Request an invite, say what you are building, and you will hear back by email when a seat opens.