Deploy your agents as dedicated AI specialists.

Yes. Modern agents do not behave like typical web apps.

A web app usually keeps a user's state in a database and loads the records it needs on each request. A modern agent, built to run on an agent harness like DeepAgents, Goose, Codex CLI, Claude Code, or OpenCode, is different. It works through a whole persistent filesystem that it can access to read/write files, store memories, etc. It needs an isolated computer to write and execute software code, access external systems and maintain security constraints.

An agent's filesystem contains source files, notes, artifacts, credentials, tool outputs, caches, and half-finished work. Serializing it into database rows on every turn would be too slow and too expensive.

Multiple simultaneous sessions can touch the same instance, so the platform has to coordinate file changes and merges. And keeping every instance's compute running forever would waste money, so Karta has to suspend, resume, meter, isolate, and route access without losing the persistent workspace.

We call each running agent computer a karta.

"karta" means "the doer" in Sanskrit.

A karta is a durable computer running a copy of the agent. It has its own filesystem, sessions, identity, tool access, metering, logs, and spend controls.

Our platform creates a karta by initializing a virtual machine with a persistent disk, installing the agent's dependencies, including the selected harness, and copying the agent definition.

The question is what each karta represents. Depending on your use case, it can be scoped to one user, one virtual employee or role in a team-owned roster, a backend job, or a fleet member.

For teams, the grain is the virtual employee, not the team itself: a team can own a roster of kartas for different roles, workstreams, business areas, or review angles.

The relationship stays yours. You own the agent project, the user relationship, and the work product. Karta runs the durable computers, isolation, metering, spend controls, and access surfaces.

Yes. Karta runs production agent fleets with enterprise-grade controls.

A prototype agent can be built quickly. The hard part starts when each user-facing assistant, each virtual employee in a team-owned roster, or each backend job needs a persistent computer with files, credentials, tool access, sessions, logs, budgets, network policy, release history, and tenant isolation. Karta handles that runtime layer: packaging agents, provisioning isolated kartas, suspending and resuming compute, coordinating filesystem state, metering usage, routing embedded access, and supporting BYOC data planes. Teams get agents to market faster and scale fleets without rebuilding security, governance, and operations around every agent.