OtoDock Platform Guide
You are an agent running on OtoDock — a self-hosted platform for teams of AI agents.
This skill is your manual for the platform itself, so you can help users set it up, run
it, and find things in it. Answer from here instead of guessing; when a question goes
deeper than this file, read the matching reference in references/.
The mental model (30 seconds)
- The server runs everything: agents, files, memory, schedules, tools, and the web
dashboard users drive it from.
- Agents are configured workers — persona + files + tools + memory + settings. Users
chat with them, schedule them, and let them collaborate.
- AI engines power agents' thinking: Claude Code and Codex (users connect
their own Claude/ChatGPT subscriptions or API keys) plus a lightweight direct
engine for low-latency work like phone calls. No model ships with the platform.
- Sessions are live running instances of an agent (a chat, a task run, a call).
The platform warms, resumes, and idles them automatically.
- Tools (MCPs) give agents abilities; skills (like this one) give technique.
- Agents run inside a strict sandbox on the server — or with full access on a
paired remote machine the user owns.
The two role systems — always state the required role
The dashboard is role-gated: users simply don't see what they can't use. So when you
point someone somewhere, say which role it needs — that's the difference between
help and confusion. Two independent systems:
Platform roles (whole installation): admin → runs the platform (all admin pages,
engines, users, MCP installs); creator → additionally creates/installs agents and
manages their own departments; member → uses the agents they've been given.
Per-agent roles (each agent separately): manager → configures the agent
(persona, tools, skills, settings); editor → works in the shared workspace;
viewer → chats, with a private personal space. A person can be manager of one agent
and viewer of another. Platform admins override per-agent roles.
Common surprises worth pre-empting: department assignment and shared-knowledge wiring
need platform admin/creator (agent managers alone can't); agent-scope schedules need
per-agent editor+; installing anything (tools, skills, agents from the community
catalog) is admin — managers and agents file requests admins approve.
Where things live (summary — full map in references/settings-map.md)
- User Settings (avatar menu): General (profile, security/2FA, appearance, wake
word, memory) · Integrations (connected accounts, API keys) · Remote Machines · AI
Engines (personal subscriptions) · Audio · Usage.
- Agent Settings (per agent): Overview · MCPs · Skills · Configuration (persona
file, engines/model, visibility mode, department, delegation targets, shared
knowledge, memory) · Monitoring (scheduled tasks, triggers, notifications, meetings).
- Admin (admins only): Users · Usage · MCP Servers · Skills · MCP Requests · Remote
Machines · Monitoring · Setup (tabs: General · AI Engines · OtoDock · Audio ·
Phone · Security · System Settings).
- The Agents page (agent pill in the top bar) is the company view: a 3D map of
departments and agents, a grid, and the community-agent browser.
What you can do yourself vs. hand to the user
You (an agent) can, with the right session role: create and manage tasks, triggers, and
notifications; browse the tool/skill catalogs and enable or request them; update your
own persona; manage knowledge libraries and department assignment via your
self-configuration tools (each change confirmed in chat); build dashboards; delegate to
wired agents; and read other agents' activity your user could see.
You cannot: install catalog packages (admin approves your request), pair machines,
manage users, change platform settings, or connect engines/accounts — for those, give
the user the exact page, tab, and role from references/settings-map.md.
Routing — read the reference that matches
| Question is about |
Read |
| Installing, first run, connecting Claude/ChatGPT/keys, voice & phone add-on, pairing remote machines |
references/setup.md |
| What an agent is, folders/workspaces, visibility modes, engines & models per agent, tools, skills, creating agents |
references/agents.md |
| Departments, the company map, delegation, meetings, shared knowledge libraries, bulletins, memory — running a company on OtoDock |
references/company-management.md |
| Schedules, one-time and trigger-fired tasks, model pinning, webhooks, event subscriptions, notifications |
references/automation.md |
| "Where is the setting for X?" / "who can do X?" |
references/settings-map.md |
House rules for platform help
- Name the place exactly — page → tab → control, with the role in parentheses:
"Under Setup → Audio (admin), add a Deepgram key…".
- Check before promising: features can be switched off per install (remote machines,
interactive terminals, chat audio, user-paired machines) — if a user can't see what
you describe, the feature may be disabled or their role too low; say which.
- Voice, phone, and live voice conversations need the phone service add-on; plain
dictation and read-aloud don't.
- Never ask users for secrets (API keys, tokens, passwords) in chat — point them at
the settings page that stores credentials encrypted.
- The public docs live at https://docs.otodock.io — link a page when the user wants
the long-form read, and fetch a page from there yourself when you need detail beyond
this skill (the site covers every feature; this skill carries the operating model and
the settings map).
1---2name: otodock-platform3description: How the OtoDock platform itself works and where every setting lives — installing and first-run setup, agents, AI engines, sessions, departments, shared knowledge libraries, tasks/triggers/notifications, remote machines, voice, and the role-labeled settings map. Use when helping a user set up, configure, administer, troubleshoot, or navigate OtoDock.4---56# OtoDock Platform Guide78You are an agent **running on OtoDock** — a self-hosted platform for teams of AI agents.9This skill is your manual for the platform itself, so you can help users set it up, run10it, and find things in it. Answer from here instead of guessing; when a question goes11deeper than this file, read the matching reference in `references/`.1213## The mental model (30 seconds)1415- **The server** runs everything: agents, files, memory, schedules, tools, and the web16 **dashboard** users drive it from.17- **Agents** are configured workers — persona + files + tools + memory + settings. Users18 chat with them, schedule them, and let them collaborate.19- **AI engines** power agents' thinking: **Claude Code** and **Codex** (users connect20 their own Claude/ChatGPT subscriptions or API keys) plus a lightweight **direct**21 engine for low-latency work like phone calls. No model ships with the platform.22- **Sessions** are live running instances of an agent (a chat, a task run, a call).23 The platform warms, resumes, and idles them automatically.24- **Tools (MCPs)** give agents abilities; **skills** (like this one) give technique.25- Agents run inside a strict **sandbox** on the server — or with full access on a26 paired **remote machine** the user owns.2728## The two role systems — always state the required role2930The dashboard is role-gated: users simply don't see what they can't use. So when you31point someone somewhere, **say which role it needs** — that's the difference between32help and confusion. Two independent systems:3334**Platform roles** (whole installation): **admin** → runs the platform (all admin pages,35engines, users, MCP installs); **creator** → additionally creates/installs agents and36manages their own departments; **member** → uses the agents they've been given.3738**Per-agent roles** (each agent separately): **manager** → configures the agent39(persona, tools, skills, settings); **editor** → works in the shared workspace;40**viewer** → chats, with a private personal space. A person can be manager of one agent41and viewer of another. Platform admins override per-agent roles.4243Common surprises worth pre-empting: department assignment and shared-knowledge wiring44need platform **admin/creator** (agent managers alone can't); agent-scope schedules need45per-agent **editor+**; installing anything (tools, skills, agents from the community46catalog) is **admin** — managers and agents file requests admins approve.4748## Where things live (summary — full map in references/settings-map.md)4950- **User Settings** (avatar menu): General (profile, security/2FA, appearance, wake51 word, memory) · Integrations (connected accounts, API keys) · Remote Machines · AI52 Engines (personal subscriptions) · Audio · Usage.53- **Agent Settings** (per agent): Overview · MCPs · Skills · Configuration (persona54 file, engines/model, visibility mode, department, delegation targets, shared55 knowledge, memory) · Monitoring (scheduled tasks, triggers, notifications, meetings).56- **Admin** (admins only): Users · Usage · MCP Servers · Skills · MCP Requests · Remote57 Machines · Monitoring · **Setup** (tabs: General · AI Engines · OtoDock · Audio ·58 Phone · Security · System Settings).59- The **Agents page** (agent pill in the top bar) is the company view: a 3D map of60 departments and agents, a grid, and the community-agent browser.6162## What you can do yourself vs. hand to the user6364You (an agent) can, with the right session role: create and manage tasks, triggers, and65notifications; browse the tool/skill catalogs and enable or request them; update your66own persona; manage knowledge libraries and department assignment via your67self-configuration tools (each change confirmed in chat); build dashboards; delegate to68wired agents; and read other agents' activity your user could see.6970You cannot: install catalog packages (admin approves your request), pair machines,71manage users, change platform settings, or connect engines/accounts — for those, give72the user the exact page, tab, and role from `references/settings-map.md`.7374## Routing — read the reference that matches7576| Question is about | Read |77| --- | --- |78| Installing, first run, connecting Claude/ChatGPT/keys, voice & phone add-on, pairing remote machines | `references/setup.md` |79| What an agent is, folders/workspaces, visibility modes, engines & models per agent, tools, skills, creating agents | `references/agents.md` |80| Departments, the company map, delegation, meetings, shared knowledge libraries, bulletins, memory — running a company on OtoDock | `references/company-management.md` |81| Schedules, one-time and trigger-fired tasks, model pinning, webhooks, event subscriptions, notifications | `references/automation.md` |82| "Where is the setting for X?" / "who can do X?" | `references/settings-map.md` |8384## House rules for platform help8586- **Name the place exactly** — page → tab → control, with the role in parentheses:87 "Under **Setup → Audio** *(admin)*, add a Deepgram key…".88- **Check before promising**: features can be switched off per install (remote machines,89 interactive terminals, chat audio, user-paired machines) — if a user can't see what90 you describe, the feature may be disabled or their role too low; say which.91- **Voice, phone, and live voice conversations** need the phone service add-on; plain92 dictation and read-aloud don't.93- **Never ask users for secrets** (API keys, tokens, passwords) in chat — point them at94 the settings page that stores credentials encrypted.95- The public docs live at **https://docs.otodock.io** — link a page when the user wants96 the long-form read, and fetch a page from there yourself when you need detail beyond97 this skill (the site covers every feature; this skill carries the operating model and98 the settings map).
OtoDock