One key, one workspace
A key is bound to a workspace at creation, so nothing in a request selects one. Read and write are granted separately, and keys are created, replaced and revoked from the platform — never over the wire.
Reach Zefi from the systems you already run. One key, scoped to one workspace, and every answer arrives attached to the account and the person it came from.
A REST API for the things Zefi does, versioned in the path and authenticated with a workspace-scoped key. It speaks JSON, returns problem documents when something is wrong, and tells you what is left of your rate limit on every response.
It exists so feedback can be collected by the systems that already know who to ask — a CRM at the moment a deal closes, a support desk when a ticket resolves, a nightly job walking a segment — instead of by a person exporting a list.
Survey links. The links Zefi would otherwise generate from the platform UI, created by your systems and carrying your own customer and respondent details.
The point is attribution. A link minted through the API knows the account and the individual you are asking, so the response arrives already attached to them rather than anonymous and waiting to be reconciled. That identity is fixed when the link is created and cannot be edited by whoever opens it.
Conversation metadata. The facts about a conversation that live in another system (the booking, the queue, the hold time), sent in daily batches keyed by your own conversation id, so they appear on the conversation and can be used to filter and score it. Only the fields agreed for your workspace are kept.
Data export. Not an endpoint, but worth knowing about: your workspace's feedback delivered on a schedule as Parquet files to an S3 bucket, ready to load into your own warehouse.
More resources will follow. Everything under Core concepts is written to hold for all of them.
| Quickstart | Your first authenticated call, in about five minutes |
| Authentication | Creating a key, choosing scopes, replacing and revoking |
| Requests and responses | Base URL, versioning, pagination, timestamps |
| Errors | The shape of a failure and which codes are worth retrying |
| Rate limits | The budgets, and how to stay inside them |
| Survey links | How links behave, and creating them in bulk |
| Conversation metadata | Sending your own data about conversations, keyed by your id |
| MCP server | Connect Zefi to Claude, ChatGPT, Cursor and other assistants |
| Export to S3 | Scheduled Parquet exports of your feedback, and how to receive them |
| API reference | Every endpoint, field and response |
v1 is current. Additive changes — new fields, new endpoints — will not break a working integration. See Versioning.