OtoDock runs Claude Code and Codex agents across departments on your server

Why it matters

OtoDock is a founder's bet that coding agents will become shared company infrastructure, with persistent work, permissions and customer-controlled deployment replacing disposable terminal sessions.

Dimitris Mourtzis highlighted OtoDock 1.6.0 on September 9th, extending his self-hosted platform for running persistent AI agents across a company. The software puts Claude Code, Codex and local models behind a shared dashboard, where agents can be assigned to departments, given recurring work and allowed to delegate tasks.

Mourtzis described OtoDock in a September 9th post as a system he built and uses to run his own business. Version 1.6.0 adds broader local-model support for agents running on paired computers and expands OtoDock's direct model API engine. It follows version 1.5.0, released on September 3rd, which added departments, remote-machine pairing, phone calls and a three-dimensional map of a company's agents.

The release cadence explains why Mourtzis is presenting OtoDock as an operating layer rather than another chat interface. Agents remain active as processes on a server, retain their own workspaces and memory, and can wake on a schedule or in response to a webhook. A marketing agent can work from shared files, a systems administrator can connect to infrastructure tools, and a department head can pass work to specialist agents beneath it.

Building beyond the terminal

Mourtzis' origin story starts with a constraint he encountered while using Anthropic's coding agent. "I wanted to use it beyond the terminal and connect it to my services," he wrote in the project's GitHub README.

He built OtoDock around that requirement, then began using the platform as the working environment for his own agents. Mourtzis says large portions of OtoDock were built, tested and shipped by agents running inside OtoDock. That is a founder claim rather than a measured development statistic, though the repository provides the code and commit history for users to inspect.

The resulting product treats an agent as a configurable worker with a persona, memory, knowledge, skills, tools and a workspace. Administrators decide which employees can use each agent and whether its history and files remain private, become shared across a team, or combine personal and common workspaces.

Each user can connect an Anthropic or ChatGPT subscription, while administrators can also configure provider API keys or local models. OtoDock runs server-side Claude Code and Codex sessions inside per-session sandboxes and streams their work to the browser dashboard. The standard installation requires a Linux server, Docker and at least 4 GB of memory, according to the installation documentation.

OtoDock also lets users pair Linux, macOS or Windows computers through an outbound connection. An agent can then operate on that machine with the files, applications, browser session and network access granted to it. If the computer goes offline, OtoDock says the server can take over the session.

That design turns a coding agent into a shared operational service. It also raises the stakes for permissions: an agent paired with a workstation can have far broader access than one confined to a server workspace.

An org chart with running processes

OtoDock enters a growing category of products trying to manage agents as organizations rather than isolated conversations. Paperclip organizes agents around goals, budgets, reporting lines and governance. Sim offers a workspace for building and governing agents, with shared knowledge, integrations, logs and controls over access and spending.

Mourtzis argues that OtoDock's distinction is its control over the runtime itself. In his September 9th post, he said Paperclip handles organization and orchestration while relying on other runtimes to execute agents. OtoDock launches Claude Code and Codex as persistent processes inside the platform, either in a server sandbox or on a paired machine.

The category boundaries are already blurring. Paperclip also supports Claude Code, Codex and several other agent types, while Sim can run in a customer's cloud. OtoDock's practical pitch rests on how much it bundles into one installation: execution, persistent workspaces, departmental permissions, scheduled work, delegation, document tools, remote computers and telephony.

The phone feature grew from a full-duplex pipeline Mourtzis built for Asterisk and FreePBX. Agents can also use Twilio to answer or place calls. Mourtzis said he personally uses voice for quick requests and questions, while relying on dictation or text for substantive work. OtoDock does not automatically route requests to the appropriate agent; users start a session with a chosen agent, which can delegate work through a configured department structure.

Self-hosted, with a commercial license

OtoDock is distributed under the Functional Source License 1.1 with a future Apache 2.0 grant. Users can run, modify and redistribute a release, but the license bars using it to compete commercially with OtoDock. Each version converts to Apache 2.0 two years after its release.

Self-hosting is free for up to five users. The pricing page lists a paid tier for four to 15 users at 25 euros per month, producing an overlap for teams with four or five seats. Larger listed plans cost 69 euros per month for 16 to 50 users and 199 euros for 51 to 100 users, before VAT.

The business model lets Mourtzis publish the full source while charging growing teams for seats, support and enterprise options. Customers provide the server and model access, leaving OtoDock to sell the management layer rather than bundle inference into its subscription.

Security becomes the product

OtoDock says every server-side agent runs in a kernel-level sandbox with restricted files, isolated processes and blocked access to private networks unless an administrator grants a specific connection. Credentials are encrypted at rest and injected only into the sessions that require them, according to OtoDock's security documentation.

The public changelog also records security fixes alongside product additions, including tighter controls for unattended tasks, phone sessions, file paths and delegated work. These are OtoDock's architectural and remediation claims, rather than independent security results. The code, threat boundaries and update history are available for prospective users to examine before giving an agent access to internal services or employee computers.

The repository showed 31 stars, three forks and 43 commits at publication, placing OtoDock at an early stage despite its broad feature list. Mourtzis has shipped quickly: the public launch announcement came on July 8th, version 1.4.0 followed on July 26th, and versions 1.5.0 and 1.6.0 arrived six days apart in September.

That pace has produced a product spanning chat, automation, documents, voice, remote computers and company-wide agent management. OtoDock's next test is whether teams want all of those capabilities under one self-hosted control plane, and whether Mourtzis can make an ambitious surface area dependable enough to become part of daily operations.