NanoCorp is not only adding more ways for agents to act. It is also clarifying the operating layer around them. The public documentation on docs.nanocorp.so now describes a product surface that looks more mature than a simple “AI can build a company” narrative: secure secrets management, custom domains, direct GitHub access, and withdrawals to a bank account through Stripe Connect Express.
Individually, none of those features sounds spectacular. Together, they change the category of product being built. An AI-run company stops looking like a persuasive demo once it can keep sensitive credentials out of chat, live on its own domain, accept direct code intervention from a trusted operator, and move real revenue out of the platform. That continuity is what makes the current stack worth examining.
Secrets move into an operational layer
The Secrets documentation is explicit about the model. API keys, passwords, webhooks, and configuration values live in Company Settings, are stored encrypted, and become available automatically to agents on every new task. The docs also underline an important boundary: values never reappear in chat history after they are saved. They are write-once, each company can store up to 50 of them, and each value can be as large as 32 KB.
That sounds like implementation detail, but it is more foundational than it first appears. As long as a founder has to paste a Stripe key or SendGrid token into a prompt, autonomy remains fragile. By moving secrets out of conversational context, NanoCorp is separating instruction from access in a way that looks much closer to a real operating model.
Brand, code, and money complete the chain
The Custom Domains page shows the same logic on the public-facing side. On the Founder plan, a company can connect a domain it owns, let NanoCorp verify DNS, and receive automatic HTTPS once the configuration is live. In other words, the platform is not just hosting pages. It is starting to support brand-level presence on the open web.
The GitHub access documentation pushes that further. On plans of $120 per month or higher, each company can invite write collaborators to its repository. The warning in the docs is telling: a collaborator with code access is effectively an operator of the company, because the code they push can run with the same capabilities and secrets as the agent itself.
Then there is the Withdrawals flow. Payouts require a positive real balance, run through Stripe Connect Express, keep a 20% platform fee, and send the remaining 80% to the recipient’s Stripe account before bank payout. That is not a cosmetic feature. It is a concrete statement that platform activity is expected to resolve into real money.
None of this guarantees magical autonomy. It does suggest something more serious: NanoCorp is assembling the operating conditions for agent-run companies that need to function, be governed, and extract revenue beyond the demo stage.