Utpal Nadiger (@utpalnadiger) launched OpenComputer Agent Deploy on July 27th, giving developers a command-line path to turn a prompt into a managed, always-on AI agent with a permanent web address.
Nadiger, a co-founder of the developer infrastructure startup Digger, described the release in a post on X as a deployment layer built on OpenComputer's durable agent sessions API. The pitch is direct: developers define what an agent should do, create its prompt and configuration files, then run oc agent deploy
to publish it. OpenComputer returns a live agent URL and a dashboard link. (opencomputer.dev) The launch moves OpenComputer above the raw compute layer it introduced earlier this year. OpenComputer provides persistent Linux virtual machines for AI agents, including root access, KVM isolation, hibernation, checkpoints and runtime resource resizing. Agent Deploy packages those primitives into a deployment workflow aimed at developers who need an agent to keep running after a coding session or local terminal closes. (opencomputer.dev)
A deployment layer for prompts
The initial workflow supports agents developed with Claude Code, Codex, opencode and Cursor. After installing the oc
command-line tool, a developer logs in, runs oc agent init
, writes the agent's instructions in prompt.md
and deploys from the project directory. The same interface can invoke the resulting agent for a smoke test. (opencomputer.dev)
Underneath that workflow, OpenComputer's Durable Agent Sessions API creates reusable agent definitions and resumable sessions. Applications can stream an agent's event log, reconnect without losing events, send instructions while it is running and receive signed webhook deliveries. OpenComputer says crashed runtimes restart automatically, while idle sessions can hibernate and wake when another message arrives. (docs.opencomputer.dev)
Each successful deployment creates an immutable revision containing the prompt, model and compatible skills. Developers can stage revisions, deploy from a linked GitHub repository and roll a built-in agent back by changing which revision is active. Running sessions remain pinned to the revision on which they started, while new sessions receive the newly activated version. (docs.opencomputer.dev)
That revision model gives prompts some of the release controls normally applied to application code. It also exposes the boundaries of the current product. OpenComputer's documentation labels durable agent sessions as a preview. Deployments containing MCP server configurations or custom runtime directories are currently rejected, and skills are not yet supported for Codex agents. (docs.opencomputer.dev)
Digger expands beyond infrastructure-as-code
OpenComputer comes from the founders of Digger, whose earlier work centered on Terraform and OpenTofu automation. CEO Igor Zalutski and CTO Mohamed Habib started Digger after working around infrastructure systems at companies including Palantir, Amazon and Fitbit. Nadiger helped launch OpenTofu, the Linux Foundation-backed Terraform fork created after HashiCorp changed Terraform's license. Initialized Capital said the founders originally built Digger's open-source Terraform runner as a side project. (joinef.com)
The same team raised a $3.6 million seed round led by Initialized Capital in June 2025. The round also included Datadog CEO Olivier Pomel, Sentry co-founder David Cramer, WorkOS CEO Michael Grinich, Resend CEO Zeno Rocha, Looker co-founder Ben Porterfield, Percona co-founder Peter Zaitsev, Motive Force Ventures and Palumni VC. Digger said at the time that its infrastructure automation software had passed 500,000 downloads and was used by more than 300 organizations, figures also cited by Initialized. (opentaco.dev)
OpenComputer applies that infrastructure background to a newer operational problem. Long-running agents need a durable filesystem, an execution boundary, credential handling, logs and a way for humans or external systems to intervene. Developers otherwise have to assemble those pieces from virtual machines, queues, webhooks and orchestration code before the agent performs its actual job.
Agent Deploy is Digger's attempt to collapse that work into a managed primitive. The comparison Nadiger made to Stripe is ambitious: Stripe standardized a complicated payments workflow behind APIs, while OpenComputer wants to do the same for an agent loop. OpenComputer has not published customer, revenue or Agent Deploy usage figures, leaving adoption as the test of whether developers want a dedicated deployment layer rather than assembling one from their existing cloud stack.
The underlying OpenComputer repository is available under the Apache 2.0 license and had 441 GitHub stars as of July 27th. The commercial service separately charges for VM compute; its published base configuration of 4 GB of memory and one virtual CPU costs $0.004 per minute while running. OpenComputer has not published separate pricing for the managed Agent Deploy layer. (github.com)