Approve everything.
Give the agent broad access and hope. Every consequential action rides on a prompt and good intentions, and you find out about mistakes after they happen.
Who Substrate is for
Built for the operator whose agents already act.
Your agents read your context, draft your messages, and touch real tools. What’s missing is the boundary: a reliable way to say what they may and may not do.
The early-access program is hosted: one invited operator per isolated environment, working closely with us. The permission product is being built and is not released.
The gap
Two bad options, and the one being built.
Agents can already act. Nothing reliably governs what they may do — so today every operator picks between two failure modes.
Approve nothing.
Gate every action behind you. The agent stays safe — and stops being an agent. You have rebuilt your own inbox with extra steps.
Write the rule once.
Substrate is being built so a rule written once, in plain language, keeps applying the same way — and stops applying when the fact behind it changes.
Working with us
What early access actually involves.
Authorize your own context.
One operator, one isolated environment, the tools and history you personally authorize. No shared workspace, no team tenancy.
Evaluate the memory today.
Substrate turns authorized context into searchable, cited memory. Ask across it, inspect the citations, and correct what is wrong — that loop is implemented now.
Define the rules that matter.
The permission product — standing rules, allow / deny / escalate, cited decision records — is being built with early operators. Your judgment on what a right decision looks like shapes it.
Fit
A strong early operator.
- You already run agents — through Hermes or a comparable stack — that take real actions on your behalf.
- You can authorize your own tools and context, and define a boundary you are accountable for.
- You can judge whether a specific authorization decision was right, and say why it was or wasn’t.
Early access