We're building the
AI agent generator.
Envelope exists because most people who've been told to "do more with AI" don't know where to start. They have a blank chat box and no structure for thinking through who does what, which tools are needed, or what to hand to an engineer. We think that's a problem worth fixing at the design layer — with natural language and a structured output you can act on.
What we believe
The best AI infrastructure is infrastructure you can see, audit, and move. When your agent definitions live only inside a conversation thread or a proprietary platform, you can't version them, you can't diff them, and you can't hand them to an engineer who needs to understand what they're building.
We built Envelope around a single idea: the output of designing a AI agent system should be a structured document — something you can read, share, store in Git, and act on. The schema is open (Apache 2.0). Any runtime can implement it. The design belongs to you.
We also believe human oversight isn't a limitation to engineer around — it's a feature to design for. Roles, reporting lines, approval gates, and tool permissions aren't bolt-ons; they're first-class concepts in the Envelope schema because AI agents that touch real systems need real guardrails from the start.
How it works
Describe what you want to automate
Tell Envelope the problem — or enter your company URL and it'll suggest ideas by department. No blank page, no YAML.
Design the team
Envelope builds out the structure — roles, tools, reporting lines, approval gates. Refine it in conversation until it matches what you actually want.
Export or hand it off
Download the spec as JSON, share a read-only link with your team, or hand it to IT with everything they need to implement it — on any conforming runtime.
Open standard, not open washing
The Envelope schema is published under Apache 2.0. The $id is https://schema.openenvelope.org/team/v1.json. Any platform can implement a conforming runtime without asking us. Validators, linters, and code generators can be built on top of it. We think the schema is the moat — not the network effect of keeping it closed.
We're building the design layer first. The runtime will follow — either ours or the community's. We'd rather have a thousand conforming runtimes than one locked platform.
The design tool and Team plan are commercial products. That's how we fund development. But the format itself belongs to the community.
The team
Ash Conway
Founder & CEO
Orchestration platforms started treating agents like employees — roles, tasks, reporting lines. The model made sense; getting started meant configuring runtimes and infrastructure most teams hadn't built. AI chat tools offered the opposite: a blank canvas with no structure for what to build or how to hand it off. Ash built Envelope as the missing layer: guardrailed design that structures your workflow, skills, tools, models, and roles — then generates the AI agents. He founded Bugwolf in 2013 after a decade in enterprise software and hardware — Cisco, SAP, Dell — and bootstrapped it to multimillion-dollar revenue.
Where we are
Envelope is in active development. The design canvas — roles, tools, reporting lines, export, and share — is live and free to use. Version history, GitHub publish, and team collaboration are coming with the Pro and Team plans.
Get in touch
Questions, partnerships, press, or just want to talk AI agents.