Hi Vapi Support, I’m building squads which needs to connect to people through business lines who don't have extensions or direct lines. The issue is that a lot have **D**ial **B**y **N**ame **D**irectories. These DBND systems are inconsistent, so the agent needs to try multiple input patterns and remember what worked for that phone system. Examples we see: - Sometimes DBND only works by FIRST name (not last name), even if the IVR says to use last name. - Sometimes you must type the first 3 letters + “#” for DBND search (but the IVR leaves out the "#"). - Sometimes you must type the FULL last name (not 3 letters). - Sometimes the person isn’t listed in DBND, so the agent should back out to the main menu and try the next best department, then the next, until options are exhausted. What I need is a reliable way to implement: 1. Memory during a call (what we tried, what worked, what failed, what menu we’re in) 2. A retry/branching strategy for DTMF + menu navigation (attempt A → fallback B → fallback C) 3. Optionally persistent memory per business/phone number (so next time we call the same number, the agent already knows the best DBND pattern and department path). I'm sure once 1 and 2 are figured out, I can just save the information in a spreadsheet and recall it. I'm sure I can write out the fallback logic (#2) so that it gets around +90% of all the nooks and crannies, but the depends on the memory (#1) actually working. That's all I need help with for now. Please let me know the options I have to implement this.

Thanks for reaching out! For guidance on capturing and storing DTMF inputs or working with API and Twilio integrations, we recommend checking out these resources: - 📘 [How to Navigate IVR Menu Systems Effectively](https://docs.vapi.ai/ivr-navigation) - 🧠 [How We Solved DTMF Reliability in Voice AI Systems](https://vapi.ai/blog/how-we-solved-dtmf-reliability-in-voice-ai-systems) These walk through setup examples, JSON formatting, and handling URL-encoded parameters step by step. If you still need help after reviewing, feel free to reply with details about your workflow, the Call ID, and we’ll point you in the right direction!

To implement memory during a call, use a custom tool (webhook) to track what has been tried and the current menu state. For example, set up a simple in-memory store on your server:

js
// Example: Store call state in memory (Node.js/Express)
const callMemory = {};
app.post('/track-attempt', (req, res) => {
  const { callId, action, result } = req.body;
  callMemory[callId] = callMemory[callId] || [];
  callMemory[callId].push({ action, result, timestamp: Date.now() });
  res.json({ status: 'ok' });
});

Call this endpoint from your Vapi assistant after each DTMF/menu attempt to keep track of the call's progress. For more, see the [IVR Navigation documentation](https://docs.vapi.ai/ivr-navigation). Source: - [IVR Navigation documentation](https://docs.vapi.ai/ivr-navigation)

This doesn’t answer my questions. It only talks about making DTMF work more reliably, not how to keep memory during the call or my other issues.