- Home
- Skills
- Agents & Orchestration
- Help a Voice Agent Remember Its Callers
More screenshots
Works with the AI Agents you already use
Help a Voice Agent Remember Its Callers
Your voice agent remembers each caller — their preferences, open issues, and facts — in one file on your own machine, no paid memory service.
$12
Help a Voice Agent Remember Its Callers
Example session with this skill installed
Extract a brief from this transcript for caller +15550199: 'Caller: I still prefer email for invoices. Also, my heater is fixed now. Agent: Great, I marked the heater issue as resolved.'
- Read your context and instructions
- Compiled the help a voice
I have processed the transcript for caller 15550199. The system performed deterministic extraction, redacted sensitive patterns, and updated the local store with one new preference and one resolved item.
Caller Brief: 15550199
| Category | Entry | Date Added | Status |
|---|---|---|---|
| Preferences | Prefers email for invoices | 2023-10-27 | Active |
| Resolved | Heater is fixed | 2023-10-27 | Expires 2023-11-26 |
| Facts | N/A | - | - |
| Open Threads | None | - | - |
Next steps
- The next call from
15550199will automatically include the email preference in the system prompt. - The "heater" issue will no longer trigger follow-up questions in scripted flows.
- Use
manual-mergeif this caller contacts you from a different number.
Connects securely to your tools. The creator never sees your data.
About this skill
A voice agent with no memory asks the same caller the same questions on every call. The paid voice platforms sell per-contact memory as an add-on; the free and self-hosted stacks have no equivalent. This skill adds the layer yourself: a small piece of middleware between your voice framework and one local store file. No cloud, no API keys, no network calls.
Inside the package: the skill file with the full design, a filled example caller brief, and a call-lifecycle checklist. Each caller gets one brief, keyed by a normalized caller ID so the same person dialing in different formats lands on one record. A brief holds four lists, and the difference between them drives everything: preferences, which never expire; open threads, the unresolved items; resolved items, date-stamped, which expire after about 30 days so briefs stay lean instead of growing into transcripts; and stable facts like plan or timezone, kept until contradicted.
A run goes like this: when a call ends, the transcript is mined for preferences, threads, and facts — rule-based, speaker-aware patterns first ("I prefer…", "still waiting…", "that's resolved"), with an optional model pass for the leftovers that can never break the call if it fails. Before anything is stored, personal data is scrubbed out by default: email addresses, phone numbers, payment card numbers (checksum-checked, so random digit strings survive), ID and account numbers, street addresses — replaced with bracketed tokens. When the next call from that person starts, a compact version of the brief goes into the agent's instructions, and scripted steps the caller already covered get marked skippable, with the reason recorded. Open threads never make a step skippable — they need follow-up, not skipping. Pipecat-style and LiveKit-style adapters are covered, both sharing the one store, and old call logs can be replayed through the same extraction to seed day-one memory.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires local file system access. Not designed for high-concurrency cloud DBs. Max characters for injection blocks are capped. PII redaction is pattern-based.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago
A Pipecat or LiveKit voice agent with call transcripts.