When a returning caller dials your AgentCall number, your AI gets a one-paragraph brief on who they are, what they last asked about, and what was promised. Every memory entry quotes the line from the transcript that produced it. Owners audit and correct anything wrong from the dashboard.
Source-backed, auditable, and entirely under your control.
Sarah called Tuesday about a quote. Friday, she calls back, and your AI already knows it is Sarah, knows about the quote, and picks up the thread. No re-asking name, no starting over.
Every memory entry quotes the verbatim line from the transcript that produced it, with the callId it came from. Review, edit, or delete any entry from the dashboard. Nothing in your AI's head you cannot trace back to a real conversation.
A Cross-call memory toggle on the dashboard overview turns it off at any time. On by default. Takes effect immediately. Never retroactively processes old calls when you flip it back on.
Four steps. All automatic. No multi-week integration.
Your virtual agent answers, or AgentCall hands the call to your own agent. The full conversation is transcribed in real time.
When the call ends, AgentCall produces a structured Call Report: who called, what they asked, what was promised, what to follow up on. Each entry is bound to a quote from the transcript.
Approved memory entries are filed against the caller. Owners can edit or delete any entry from the dashboard before or after it is used.
The moment the same caller dials back, your AI receives a one-paragraph brief plus the underlying memories. The conversation picks up where it left off.
One call. Returns the brief your AI needs to speak with full context. Use it from MCP inside Claude or Cursor, the SDK, or plain REST.
curl https://api.agentcall.co/v1/memory/next-call-context \
-H "Authorization: Bearer YOUR_API_KEY" \
-G --data-urlencode "phone=+14155550142"{
"caller": {
"name": "Sarah Chen",
"phone": "+14155550142",
"firstSeenAt": "2026-05-12T14:03:21Z",
"callCount": 3
},
"brief": "Sarah called Tuesday asking about pricing for a 30-person team. You promised to send a quote by Friday. She mentioned a competitor demo on Thursday.",
"memories": [
{
"id": "mem_8h2k",
"fact": "Team size is approximately 30 people",
"source": "Yeah we're a team of about 30",
"fromCallId": "call_x7k9m2",
"recordedAt": "2026-05-12T14:05:11Z"
},
{
"id": "mem_9p4n",
"fact": "Promised a written quote by Friday",
"source": "Can you send a quote by Friday?",
"fromCallId": "call_x7k9m2",
"recordedAt": "2026-05-12T14:06:48Z"
}
]
}The same data is available via the get_next_call_context MCP tool. Full payload and event types at /docs/memory.
Auditable Call Memory works with any AI phone agent. Pick the pattern that matches how your agent is built.
Drop the AgentCall MCP server into your client. 60 tools and 5 prompts auto-discover, including get_next_call_context, list_contacts, get_memory_timeline, and approve_memory_candidate. Your agent reads memory and acts on it from the same surface.
{
"mcpServers": {
"agentcall": {
"transport": "streamable-http",
"url": "https://api.agentcall.co/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}Use the pre-call context webhook to inject memory at the start of every inbound call, and the post-call transcript webhook to feed fresh Call Reports back into your platform. HMAC-signed, retried with backoff. Open-source Cloudflare Worker template available.
Configure inbound AI on any AgentCall number, write your system prompt, and memory is wired in automatically. Returning callers get recognized and the brief is loaded before the AI speaks. No code required.
Wire the pre-call context webhook to your agent platform so the brief gets composed by your own brain. Memory is still extracted and stored by AgentCall; your platform decides which pieces to surface on the next call. Hermes setup guide.
Every example below is shipping today on AgentCall.
Your customer calls Tuesday about a quote. They call again Friday. Your AI greets them by context: 'Hi Sarah, are you calling back about the quote we discussed Tuesday?' No re-introduction, no re-asking what it is about.
Each Call Report feeds your CRM. Your AI reads prior conversation history before dialing the next prospect, references what was already discussed, and never restarts qualification from scratch.
Customer calls Monday and gets routed to your AI for triage. Calls back Tuesday. Your agent already knows the issue, the steps tried, and what to try next.
Your AI agent makes follow-up calls based on outcomes from the prior round. Anyone who said 'call me Tuesday' gets a Tuesday call referencing the previous conversation.
No separate memory charge. Memory is included on Pro, and Free is capped naturally by the inbound AI trial.
Free tier includes one US local number, 5 minutes of inbound AI voice per month, and the dashboard audit view. No card required to start.
Or read the docs: Auditable Call Memory · Post-call Transcript Webhook · MCP Server · FAQ
Auditable Call Memory is a source-backed memory layer extracted from completed AI calls. When a returning caller dials your AgentCall number, your AI receives a one-paragraph brief with who they are, what they last asked about, and what was promised. Every memory entry quotes the transcript line that produced it so owners can audit and correct anything wrong from the dashboard.
It works with any AI phone agent. AgentCall's own virtual agent uses it out of the box. You can also pull memory into Claude, Cursor, Windsurf, OpenClaw, or your own agent platform via the MCP server, the pre-call context webhook, or the REST API. Hermes is one supported brain among many.
There is a Cross-call memory toggle on the dashboard overview. Turning it off takes effect immediately. It is on by default for new accounts and never retroactively processes past calls.
Each entry has a subject (the caller or topic), a short fact, a source quote pulled directly from the transcript, the callId it came from, and a timestamp. Owners can edit or delete any entry from the dashboard.
There is no separate charge for memory on the Pro plan ($19.99/month). The Free plan is naturally capped by the 5-minute monthly inbound AI voice trial, so memory only builds up to the limit of what your free trial covers.