Tools & Connectors

Tools are capabilities your AI can call — beyond just answering. Connectors are secure one-click integrations with the services you already use. Together they turn your AI from a talker into a doer.

Where to manage them

• Built-in tools — enabled per persona from the persona's settings; most are on by default.
• Custom tools — create them in Works → AI Library (Tools category), then attach to personas.
• Connectors — connect services in Settings → Connectors (one-click OAuth), then enable per persona.
• Custom MCP servers — also under Settings → Connectors.

Key features

• Built-in tools — knowledge search, web, code, file generation, and memory work out of the box, no setup.
• Custom tools without servers — template tools orchestrate built-in capabilities; webhook tools call your own endpoint.
• One-click connectors — OAuth integrations with Google, Slack, GitHub, Notion, and 20+ more services.
• Per-persona control — every tool and connector is enabled per AI, with allowlist and denylist patterns.
• Bring your own MCP server — add a custom remote MCP server by URL.

When Tyrex uses tools

Your AI calls a tool when your request maps to a tool's described capability — either explicitly ("search the web for the latest version") or implicitly ("is there anything new in my Drive?" with Drive connected).

Tyrex does not call a tool for general knowledge questions about a connected service. Asking "how do Notion databases work?" with Notion connected is answered directly; asking "what's in my Projects database?" triggers the tool.

You can steer how readily your AI reaches for tools through instructions — see the persona's always-on instructions in your AI's settings.

Built-in tools

Every Tyrex AI comes with a set of native tools:

Your own tools

There are two kinds of custom tools, and the schema is exact — Tyrex validates tools against it:

Template tools

A template tool is configuration plus orchestration instructions. When the AI calls it, Tyrex runs the primitives you specified in the order your instructions describe.

{
  "name": "make_questions",
  "description": "Create practice questions from the attached knowledge",
  "parameters": {
    "type": "object",
    "properties": {
      "topic": { "type": "string", "description": "The topic to cover" },
      "count": { "type": "integer", "description": "How many questions" }
    },
    "required": ["topic"]
  },
  "primitives": ["rag_query", "create_artifact"]
}

Plus an instruction telling the AI how to compose the steps:

Search the attached knowledge for the requested topic and count.
Draft questions at the requested difficulty.
Save them as a PDF artifact and return the download link.

Template tool fields

Available primitives

Primitives are resolved against the live tool registry at run time — referencing a name that doesn't exist returns a clear error in the run, so you can test and adjust.

Webhook tools

A webhook tool forwards the model's arguments to an endpoint you control. Your endpoint receives the arguments as JSON and returns a result the AI can use.

{
  "name": "check_stock",
  "description": "Check product stock at the store",
  "parameters": {
    "type": "object",
    "properties": {
      "sku": { "type": "string", "description": "Product SKU" }
    },
    "required": ["sku"]
  },
  "webhookUrl": "https://api.yourstore.com/check-stock"
}

Webhook tool fields

The request your endpoint receives

POST https://api.yourstore.com/check-stock
Content-Type: application/json

{
  "sku": "FNB-001"
}

Your endpoint should respond with JSON within the timeout. Large responses are truncated with a marker, so keep payloads compact.

Connectors

Connectors are one-click OAuth integrations with the services you use — no tokens to paste, no API keys to manage. Tyrex connects through its own OAuth proxy, so your service password is never exposed.

Available connectors include GitHub, Slack, Asana, Notion, Trello, and more from the catalog. Google connectors (Gmail, Drive, Calendar, Docs/Sheets/Slides) are marked Coming soon while their OAuth verification is in progress — they'll light up automatically once ready.

Connectors work inside workflows and chat alike — fetch a document from Drive, summarize it, and deliver the summary to Telegram, all in one workflow.

Custom MCP servers

You can add your own remote MCP server by URL. Custom connectors accept:

Local STDIO servers cannot be connected directly — only servers publicly exposed over HTTP.

Scoping tools & connectors to personas

Every tool and connector is granted per persona. This is where the allowlist and denylist patterns matter — your customer-service AI doesn't need file generation unless you say so.

Pattern 1 — Enable everything

The simplest pattern — give an AI every tool and connector you've connected:

• Leave the AI's tool list at the default (all enabled).
• The AI chooses what it needs from the request.

Pattern 2 — Allowlist: enable only specific tools

Enable only the capabilities this AI needs:

1. Turn off the AI's tools by default.
2. Enable only the tools for its job — e.g. rag_query and web_fetch for a support AI.

Every other tool stays hidden from the model — it won't be offered, and it won't be called.

Pattern 3 — Denylist: disable specific tools

Enable everything by default, then explicitly disable the tools you don't want:

• A read-only assistant that answers from your knowledge base → disable create_artifact and run_code.
• An AI that shouldn't touch your calendar → leave Calendar connected but disable it for that AI.

Denylisting write or destructive capabilities is recommended when you want a human confirmation step before state changes.

Configuration precedence

Settings resolve with this precedence (highest to lowest):

1. Per-persona toggle (the AI's own tool/connector switch)
2. Account-level default (Settings → Connectors / Tools)
3. Platform defaults

For example, with Telegram connected at the account level and disabled for your customer-service AI, that AI can never send messages — even though your other AIs can.

Validation rules

Tyrex enforces these rules on custom tools and connectors:

• name and description are required — tools without them are rejected at save time.
• parameters must be a valid JSON Schema with properties and required.
• Webhook URLs must be HTTPS and public — private networks and localhost are blocked (no server-side request forgery).
• Unknown primitives in a template tool return a clear run-time error, so you can test and adjust.
• MCP servers must be public HTTP(S) — STDIO servers are not supported.

Error handling

Custom tool failures never break the conversation — the AI sees the error and adapts:

Every tool call is logged with the conversation, so you can audit what your AI did and why.

Security model

• Connectors use standard OAuth — Tyrex never sees your service password, and access is limited to the permission scopes you grant.
• Webhook tools are validated — HTTPS only, no internal addresses, rate limited and timeout enforced.
• Per-persona scope — a tool or connector is only reachable by the AIs you enable it for.
• Full audit trail — every tool call is part of the conversation history.

FAQ

Do I need to pay to use my own webhook tools? Using your own tools with your own AIs is free. Listing them in the Marketplace is a separate step.

Can one connector serve multiple AIs? Yes — connect once at the account level, then enable it per persona with the allowlist/denylist patterns above.

My webhook endpoint is behind a VPN — can Tyrex reach it? No. Endpoints must be publicly reachable over HTTPS.

Do connectors work in workflows? Yes — the same connected services are available to workflow steps, so a workflow can read from Drive and deliver to Telegram.

Can I sell my custom tool? Yes — see Sell on the Marketplace for the listing flow, review process, and fees.

Next: Skills & Plugins.