OtoDock: Self-Hosted Multi-Tenant Platform for Claude Code and Codex Agents
OtoDock is a self-hosted platform that runs multiple AI coding agents — powered by Claude Code or Codex CLI — as persistent processes on your server. It coordinates them into what the author calls a "company OS": agents get departments, schedules, shared or private workspaces, and even phone numbers. Today's release is version 1.6.0, and everything is public under a fair-source license; self-hosting is free for up to five users.
Core architecture and sandboxing
Each agent runs on your server in a kernel sandbox (bubblewrap) with network isolation always on (pasta). Every agent gets its own workspace, memory, and schedules, and can be configured to run identically on any remote computer using a single outbound WebSocket — no inbound ports or VPN needed. You can bring your own Anthropic subscription (for Claude Code) or OpenAI subscription (for Codex), or run local models. The engine is chosen per agent and can even be switched per chat.
Agent anatomy and sharing modes
Each agent has six editable components — persona, memory, workspace, knowledge, skills, and tools. Four sharing modes control how work lands:
- Personal only: a private workspace for each person.
- Personal + shared: private default, plus a team workspace.
- Shared + personal: shared default, plus a private space.
- Shared only: one history and workspace for the whole team.
Roles are tiered: Admins run the platform, Managers control agents, Creators make them, Editors modify shared files, Members and Viewers use or read only. Multi-tenancy means many people can collaborate on the same agents without stepping on each other's work.
Built-in features
Agents aren't limited to coding. They can make and receive phone calls through Twilio or your own Asterisk server, edit and preview videos, Excel, Word, and PPT files inline via Collabora, and notify you with four severity levels. They can run on schedules, on webhook events, or on demand. Each run is a full chat that can be reopened and continued — you don't have to watch it live.
The author notes that large parts of OtoDock were written by OtoDock itself running Claude Code, and includes this challenge for HN: "I would love people to read the code, take a look in the sandbox model, and tell me what would stop you from running this on your own hardware."
📖 Read the full source: HN AI Agents
👀 See Also

RTX 5060 Ti 16GB Local LLM Benchmarks: 30B Models Still Lead for Coding
Benchmarks on an RTX 5060 Ti 16GB show Unsloth Qwen3-Coder-30B UD-Q3_K_XL achieving 76.3 tok/s on Ubuntu with quality score 8.14, making it the recommended default coding model. The Unsloth Qwen3.5-35B UD-Q2_K_XL hits 80.1 tok/s but with lower quality scores.

Prism MCP v2.1 adds persistent memory to Claude sessions
Prism MCP v2.1.0 'The Mind Palace' provides persistent session memory for Claude, eliminating the need to re-explain project context. It features local SQLite storage, a visual memory browser, state rollback, and cross-client context sync.

OpenClaw Skill Pack: A 2,500+ Command Set for Real Autonomous Operations on Ubuntu
A new skill package for OpenClaw AI agents introduces over 2,500 execution skills for DevOps operations like Docker management, networking configuration, CVE response, and system automation on Ubuntu environments.

Awesome OpenClaw Skills Repository Provides 5,400+ Filtered Skills
A GitHub repository called awesome-openclaw-skills offers 1,715+ production-ready skills that AI agents can install with one CLI command, filtered from the official OpenClaw Skills Registry.