integrations/vapi/v0.1.0
bytedance/flowgram.aiintegrations/vapi/v0.1.0Sep 21, 2026by het0814
Upgrading?
AI Summary
First release of `memanto-vapi`, a webhook integration designed to provide memory to Vapi voice agents. It enables agents to access organizational knowledge and past call learnings, while supporting optional per-caller memory and automatic lesson extraction.
Key Highlights
- Memory injection at call start with a 3-second lookup cap
- Mid-call tools for real-time memory recall and remember
- Automatic lesson extraction from end-of-call reports in the background
- Dual memory scopes: shared (organization-wide) and caller (private)
New Features
- Memory at call start (`{{memanto_context}}`)
- Mid-call tools (`memanto_recall` and `memanto_remember`)
- Automatic learning from conversation summaries
- Support for shared and caller-specific memory scopes
- FastAPI server and CLI integration (`memanto-vapi serve`)
- Bearer token authentication (legacy `X-Vapi-Secret` supported)
Full Release Notes
# Release Notes for memanto-vapi v0.1.0
First release of `memanto-vapi`: a single webhook that gives a Vapi voice agent memory between calls. The agent gets your organization's knowledge and the lessons it learned on earlier calls, with optional per-caller memory.
## New Featuresper-caller memory.
## New Features
- **Memory at call start** (`memanto_vapi/memory.py`)
- Replies to `assistant-request` with `{{memanto_context}}`, filled from recent and relevant memories.
- Lookup is capped at 3s (Vapi allows 7.5s); if it times out or fails, the call starts without memory.
- **Mid-call tools** (`memanto_vapi/tools.py`)
- `memanto_recall` and `memanto_remember` function tools. `memanto-vapi tools --server-url …` prints the JSON to create them with Vapi's `POST /tool`.
- **Learning after each call**
- On `end-of-call-report`, lessons are pulled from the conversation (mistakes, corrections, better answers, business facts) using Vapi's summary and success evaluation. This runs in the background after the webhook replies.
- A Vapi retry of the same report is skipped (`retained-<call id>` tag).
- **Two scopes** (`MEMANTO_VAPI_SCOPE`)
- `shared` (default): one memory for the whole assistant, with no caller details stored.
- `caller`: adds memory private to each caller, tagged with a salted HMAC of their phone number or `customer.externalId`. Raw numbers are never stored, and the server decides whose memory a call can read or write.
- **Server and CLI** (`memanto_vapi/webhook.py`)
- `memanto-vapi serve`, or `create_router` to mount the webhook in your own FastAPI app.
- Requests are authenticated with a bearer secret (legacy `X-Vapi-Secret` also accepted).
- `build_assistant_overrides()` covers outbound, web and chat calls.
## Tests
- 36 tests covering both scopes, caller isolation, auth, timeouts, retries and malformed payloads. Payloads follow Vapi's OpenAPI schema.
## Install
pip install memanto-vapi requires `memanto>=0.2.21`.Related Repositories
The AI that really does things. Any OS. Any Platform. The lobster way. 🦞
Interactive roadmaps, guides and other educational content to help developers grow in their careers.
Reverse engineer anything with agents, from app behavior down to native binaries.
The open source coding agent.
Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.