Authentik is the company's Single Sign-On (SSO) system — one account logs you into Loom (and other tools that have finished Authentik linking). If you already have an Authentik account, skip to Step 2.
Contact Jeff and ask him to create your Authentik account. He will send you an enrollment link that looks like:
auth.prod.wellplay.io/if/flow/default-source-enrollment/?itoken=…
Now that you have your Authentik account, log into the Loom portal. Loom is your AI control centre — it's where you manage your tool connections and generate your AI token.
NDoc, Kanban, and GitHub use OAuth — a secure one-time authorisation. Once connected, your AI assistant can read and write to these tools on your behalf.
Memory is Naton Lab's private AI memory system. Unlike shared tools, each person's memory pool is completely private — only you can read and write to it. You need a personal API key to unlock it.
Mattermost is the company's chat platform. With your personal API key, your AI assistant can read messages, send replies, and search conversations — all as you, not a shared bot.
The final step — generate your personal AI token and connect your AI tools. After this, your AI assistant will have access to all the tools you set up.
~/.cursor/mcp.json.In Claude Desktop (the app — not claude.ai in a browser): needs Node.js. Copy the Claude Desktop snippet, then fully quit & reopen — Claude logs you in via OAuth automatically.
In Claude Code / Codex: copy the relevant snippet from the AI Config page.
In Open WebUI: MCP via OAuth on
llm.os.wellplay.io; optional model brain via /v1 — see Step 7.
/connect DM anymore. Go to My Connections → Personal Bot: paste your Mattermost bot token + username, pick tool permissions and a Reply model (from AI Models), then create. Loom pushes config to Bot Server for you.
Loom’s /v1 endpoint is ready for beta use as an OpenAI-compatible model router for Cursor, Open WebUI, Mattermost Personal Bot, and scripts. Save a provider API key here once, then use your AI Config JWT as the Bearer token — not the provider key itself. For Claude Code / Desktop, Codex, and other AI clients, connect Loom as MCP only and keep that client’s own subscription for the model brain.
Ollama Cloud (hosted open models): ollama.com/settings/keys — model IDs like
ollama-cloud/gpt-oss:120bCloudflare Workers AI (token + Account ID): Workers AI
Google Gemini: aistudio.google.com/apikey
Ollama: company server — no key; needs admin
OLLAMA_BASE_URL. For local Ollama / LM Studio, point Cursor at localhost directly (not via Loom).
/v1 will work. Without this step, chat requests return 403 provider not configured. Yellow Saved — verify means fix the checklist and click Test ▶. Use Try it out on AI Models to chat through Loom first, then copy a model ID into Cursor or Open WebUI.Cursor / Open WebUI / OpenAI clients:
baseURL = https://loom.prod.wellplay.io/v1, API Key = that AI JWT. Model IDs need a provider prefix — copy from AI Models or GET /v1/models, e.g. openrouter/openrouter/free. Do not use bare openrouter/free.Claude Code / Desktop / Codex: MCP only — copy the MCP snippet from AI Config. Use that product’s own subscription for the model. Loom does not support Anthropic Messages or Codex Responses as a product path.