Let the AI call your systems, not guess about them
Define a tool once with a name, a description, parameters and your endpoint, attach it to any agent, and the model calls it mid-conversation to look up a booking, check availability or open a ticket. The reply the guest sees already contains the result.
Standards-based function calling ✦ Five-minute timeout ✦ Loop cap per turn
The most guest-obsessed brands trust Lodgestory to power 50M+ conversations a year
BBingo Hotels
Agents are per channel. Tools are for everyone.
Give WhatsApp a warm concierge voice and email a measured operations tone, then attach the same tools to both. A tool is defined once in the shared library and linked to as many agents as need it; update its URL and every agent picks up the new endpoint on its next call. Tool definitions follow the function-calling format the major AI platforms use, so moving tools in and out is straightforward.
- No hard caps. As many tools per agent and agents per channel as you like; five to ten tools per agent is the practical sweet spot.
- Channel summary. See at a glance which channels have an agent and which have none.
- Partner API. Create and update agents, tools and attachments programmatically.
Tools library
Built for answers that depend on your data
Describe it, do not script it
A name, a sentence on when to use it, typed parameters and an endpoint. The model decides when to call and what to send.
One turn, result included
Lodgestory runs the call, hands the result back to the model and waits for the reply. The guest sees typing, then an answer with the facts in it.
Bounded by design
A five-minute timeout per call and a cap on tool-call loops per turn keep a slow or broken endpoint from holding a conversation.
Three runtimes, one definition
Bot journeys, inbox drafts and answer validation all read the same agents and tools, so a change lands everywhere on the next run.
Ask, call, read, answer, inside one turn
A guest asks about a booking. The model recognises the reference, picks getBookingStatus and fills bookingRef. Lodgestory makes the request with your headers, gets the JSON back and returns it to the model, which weaves the status into a reply in the persona's voice. If a second lookup is needed, the loop runs again, up to the cap. When an endpoint fails, the agent is told in plain words and tells the guest it will check with the team.
- Arguments typed. Strings, numbers and required flags, so a malformed call never leaves Lodgestory.
- Headers for auth. Bearer tokens and API keys live on the tool, never in the prompt.
- Results stay fresh. With chat actions on, tool results are not kept in history; the agent calls again when it needs the data later.
Ready to connect the AI to your own data?
One tool and one persona take ten minutes. Test it in a journey with a real booking reference.
Questions, answered
About tools and how the AI uses them.
A named action the AI can invoke: look up a booking, check availability, fetch the weekend rate, create a ticket. A tool has a name the model uses, a description that tells it when to call, parameter definitions, and the URL, HTTP method and headers of your endpoint. Tools are organisation-wide and can be attached to any AI agent.
When an agent is invoked, Lodgestory loads its persona and attached tools, builds the conversation context and sends it to the model. If the model decides to call a tool, Lodgestory runs the HTTP call on your behalf, returns the result to the model and waits for the next response, all inside one turn. The guest sees typing, then a reply that already contains the result.
Tool calls time out at five minutes by default, so a slow or failing endpoint cannot hold the conversation, and a safety cap limits how many tool-call loops a single turn can run, so the AI cannot get stuck in a cycle. Failures are reported to the agent in plain words and it carries on.
Everywhere an AI agent runs: the AI Agent step in bot journeys, the AI Assistant drafts in the inbox, and AI-based validation of free-text answers. Agents are per channel; tools are shared, so the same getBookingStatus serves WhatsApp, email and web chat without duplication.