Changelog

What may change, and what may not

• Additions ship at any time without a new version: new endpoints, new optional request fields, new fields in an answer, new values in a field like status, and new event types. Read the fields you know and ignore the rest, and treat a value you do not recognise as "something else".
• Anything that would break a working integration goes into /v2. /v1 keeps working.
• A planned removal is announced here, and the responses it affects carry Deprecation and Sunset headers before anything is removed.

October 2026 — /v1

The first release of the API.

• Calls: place a call with an outbound agent, with its variables and your metadata, safely retried with an Idempotency-Key; end a call, or cancel one still ringing; name the hours a call may start in.
• Scheduled calls: ask for a call later, placed at its time or at the next opening of its calling hours, and cancel it before then.
• Campaigns: call a list of people at a pace you set, in calling hours read in each person's own time zone; start, pause, resume and cancel it; read how each person's calls went; and hear each step as an event.
• Contacts: keep each project's phonebook in step — add, change, find by number, delete, and import up to 1,000 people a request. An inbound agent can fill its fields from the caller's entry.
• Knowledge: add, replace and delete the files your agents answer from, follow their indexing, and have a website read again.
• Browser calls: start a voice call with an agent from your server and join it from your own page, with @aigently/web.
• Chats: start and continue a chat with any agent, one message at a time, as a whole reply or streamed; send pictures.
• Conversations: list, read and sync them, with transcripts, analyses and five-minute recording links; find them by what was said or by your own external_id; read what each cost; run the analysis again; delete them, and list the ones deleted.
• Inbound calls: the context lookup asks your server who is calling while the phone is answered.
• Webhooks: an event for each change in a conversation's life, signed the Standard Webhooks way, with live and test receivers. Manage receivers through the API, and rotate a secret with an overlap so no event is refused while you update.
• Events: every event your receivers were sent, kept for 30 days, to recover what one missed.
• Billing: the credit balance and every movement of credit.
• Phone numbers: list them, and point one at an agent with a live key.
• Do not call: one list per organization that no agent rings, kept from your own systems.
• Keys: limit a key to your servers' addresses. A key pushed to a public repository on GitHub is revoked, once our place in GitHub's secret scanning program is confirmed.
• Test mode: test keys, test numbers that ring nobody, test calls to your own agents, and aigently listen for your laptop.
• AI assistants: the API as an MCP server, read-only unless a key allows changes.
• Audit log: everything people and keys did in your organization, for your security system.
• Versions: list the versions an agent was published as, and pin a conversation to one.
• Agents as code: read, create, replace, publish and delete agents from their definitions.
• Tools and credentials: make, change and delete tools and MCP servers, and add tool credentials to the vault, where no answer ever carries one back.
• Evaluations: rehearse an agent's saved conversations from your build pipeline and read how each did; aigently evaluate fails the build step when one no longer passes.
• Live calls: follow a call's transcript while it happens, and have the agent say a sentence, take a note from your systems, or transfer the caller to a person.
• Mobile apps: iOS and Android SDKs that join a call your server started and run a chat, as @aigently/web does in a page.
• Terraform: keep agents, tools, credentials, evaluations and webhook receivers in Terraform.
• WhatsApp: start a WhatsApp conversation with an approved template, and the number's agent answers when the person writes back.
• The command line: every operation as a command, and an agent's fields written as TypeScript or Python types.
• Examples in seven languages for every operation, run against the platform before publishing.