{title}
-{copy}
-CHATGPT + GROUNDCONTROL
++ Connect once through OAuth. Choose the exact deployments and capabilities. GroundControl keeps the + infrastructure credentials and executes the work. +
+THE RESULT
+SELF-HOSTED CONTROL PLANE FOR SOFTWARE AGENTS
-
- GroundControl lets ChatGPT and other approved agents inspect, deploy, verify and recover
- applications on infrastructure you own without handing the conversation your SSH keys,
- provider credentials or an unrestricted shell.
+ SELF-HOSTED CONTROL PLANE FOR SOFTWARE AGENTS
+ GroundControl gives ChatGPT and other approved agents typed deployment, runtime and recovery capabilities
+ on infrastructure you own. No SSH keys in the conversation. No generic remote shell.
THE PRODUCT THESIS
- GroundControl turns repositories, containers, domains, deployment evidence and provider connections
- into typed operational capabilities. The agent asks to inspect or redeploy. GroundControl handles
- the host mechanics, verifies the result and returns evidence.
- CHATGPT + YOUR INFRASTRUCTURE THE PRODUCT THESIS
- GroundControl exposes a remote MCP endpoint with OAuth. Each client receives exact deployment and
- capability scopes. Human approval is reserved for the boundaries you choose, not every harmless operation.
+ GroundControl turns repositories, containers, domains, health and deployment evidence into bounded
+ operational capabilities. The agent says what should happen. GroundControl handles the host mechanics
+ and returns proof.
{copy} {copy} REAL OPERATING SURFACE
- The control plane keeps deployment identity, runtime state and public verification close enough that
- agents can make decisions without scraping a pile of shell output.
- Use an operator-owned GitHub App, then choose the exact account and repository GroundControl may see. Link an installation-backed repository to the deployment. Private repositories are first-class, not a public-API fallback. Operations keep their own durable state and evidence, even when the calling chat disappears. Connector health distinguishes configured, healthy, degraded, missing-scope and revoked capabilities. CURRENT PRODUCT LIVE ACCEPTANCE · SEPTEMBER 2026
+ The agent received a durable operation ID. GroundControl reconciled the deployment, checked Postgres,
+ Redis, migrations, API and web, verified the running images, then hit the public endpoint and returned
+ the evidence to ChatGPT.
+ WHY AGENTS PREFER THE CONTROL PLANE WHAT GROUNDCONTROL GIVES THE AGENT {feature.copy} {copy} CONNECTORS AS CAPABILITIES NARROW TOOL SURFACE
- GitHub repository reads, signed webhooks, repair PR permissions, GHCR pulls and Daytona sandbox
- lifecycle are verified independently. GroundControl tells agents what works now and exactly what needs repair.
+ External agents do not get the human terminal. They get intent-level operations with resource scopes,
+ idempotency and evidence.
healthy · repository read verified unverified · configured but not yet proven missing_scope · reconnect with the capability required CONNECTORS AS CAPABILITIES
+ Repository access, signed events, repair PR permissions, private GHCR pulls and Daytona sandbox
+ reproduction are verified independently. “Configured” does not get to cosplay as healthy.
+ AGENT-ASSISTED INSTALL
- The installer runs on the VPS, generates secrets there, starts GroundControl on loopback and returns a
- short-lived one-time claim. The installing agent never becomes the permanent administrator.
+ The installer runs on the authorized VPS, generates secrets there, starts GroundControl on loopback and
+ returns a short-lived one-time claim. The installing agent never becomes the permanent administrator.
{copy} {copy}
- A fresh install binds locally first. GroundControl should be published through HTTPS before you use the
- operator login over the public internet. Domain/bridge automation is the next distribution step.
- YOUR INFRASTRUCTURE, NOW AGENT-ADDRESSABLE
- The agent gets typed capabilities and evidence. You keep ownership, credentials and the private control plane.
+
+ Your agent gets bounded reach and operational evidence. You keep ownership, credentials and the private
+ control plane.
+ Give your agents
+ infrastructure arms.
+
+ Agents should ask for outcomes, not learn your server.
- Connect once. Grant only what the agent needs.
- Agents should ask for outcomes, not learn your server.
{title}
- {name}
- GroundControl knows what is actually running.
-
-
- The interface changed. The proof should change with it.
+ This is not a mockup. ChatGPT redeployed a real workload through GroundControl.
+ More reach. Less infrastructure-specific reasoning.
{feature.title}
- {title}
+ “Configured” is not the same thing as healthy.
+ Capabilities, not a bag of shell commands.
{name}
+ {copy}
+ Connect GitHub. Connect Repo. Prove what actually works.
+ Install mechanically. Claim ownership personally.
Build anywhere. Let GroundControl operate what you ship.
-