What's new. Only what actually shipped.

Every release of zerosoft, newest first. Written from the real commit history: no roadmap promises here.

11 releases
  1. Connect Claude to all 24 apps, and a zerosoft CLI

    • AgentsOne connector for Claude Code. You sign in with your normal login, choose the workspace and whether it can write, and Claude acts as you, with your role and permissions. See and disconnect your apps in Settings, API and webhooks.
    • AgentsA zerosoft command line with the same reach, for the terminal and for scripts. It runs from the repository for now and is not on npm.
  2. Agents: procedures, buttons and forms, email, WhatsApp, Slack, Messenger, Instagram and phone

    • AgentsProcedures: step-by-step flows the agent follows in order, with questions it validates (email, phone, number, buttons), forms, if-then branches, tool actions and handoff. They start with a keyword or when the agent recognizes the case, and the visitor can write “cancel” to leave.
    • AgentsButtons and forms inside the chat (widget, public page and playground). A form can save the contact in the CRM with a note; on WhatsApp, Slack and email the options arrive numbered.
    • AgentsNew channels: emails to your Chat widget addresses, new WhatsApp chats in Groups, Slack, Messenger and Instagram with your own app, and calls to a Twilio number that the agent answers out loud and can transfer.
    • AgentsKnowledge: semantic search (opt-in) finds the right passage when the question uses other words, and public Zendesk help centers come in and refresh every day.
    • AgentsTools: your own JavaScript functions in an isolated machine with no network, Stripe billing of the identified visitor and Shopify order status and products (read only, with your keys, encrypted).
    • AgentsA/B tests between a saved version and the live one, access per agent (members can edit, view or not see it), and optional AI redaction of names and addresses in transcripts.
  3. Agents: versions, files and Q&A, MCP, simulations and AI quality scores

    • AgentsPublish a version and keep editing as a draft: the playground and tests use the draft, your live channels keep the published version, and any version can go back live. The Versions tab shows who changed what.
    • AgentsKnowledge from PDF, Word, TXT, Markdown and CSV files, questions and answers with variants and CSV import, and sitemaps of up to 300 pages. Replies cite their sources with a link, and questions the agent could not answer become one click answers.
    • AgentsConnect MCP servers as tools, with the token encrypted and approval from an admin before any tool that changes data. Instructions per channel, a check for rules that collide, spam words and blocked countries or visitors.
    • AgentsWith a signed-in visitor, the agent reads their CRM contact, Cal bookings and Track requests and remembers them between conversations. Personal data can be hidden in saved transcripts.
    • AgentsSimulated conversations with an AI customer, replays of your last real conversations against the draft, and an AI score for every conversation (satisfaction, sentiment, topic and your scorecard) with a review queue, handoff reasons, results per channel and CSV export.
  4. Nested blocks and undo in Docs, workshops in Board, milestones in Track and a sidebar list in every app that needs one

    • AgentsOwners can require a teammate to approve chosen write actions before they run.

    Closes 45 ideas from the board

  5. Version history in Notes and Board, alerts in Flow and Robots, analytics in Cal and Agents

    • AgentsAn Analytics tab with resolution, handoffs, tool errors, step limits and topics, and an email to the owner when thumbs down, tool errors or step limits spike.

    Closes 38 ideas from the board

  6. Recycle bin in CRM, approvals in Cal, images in Docs and tests for Agents

    • AgentsThe playground simulates write actions, saved tests are graded by a model, http_request can send secret headers to allowed domains, and the agent list shows 7-day usage per channel including API and Flow.

    Closes 25 ideas from the board

  7. Twenty-one fixes from the backlog across ten apps

    • AgentsAfter a handoff the agent stops answering and a teammate can reply from the Conversations tab. Knowledge URLs download on save and refresh daily.

    Closes 21 ideas from the board

  8. Agents: AI agents that answer and take action

    • AgentsNew app: Agents, an open-source alternative to Intercom Fin. Give an agent instructions and knowledge (your Docs and web pages), and tools across the suite: search knowledge, save leads to the CRM, offer times and book meetings in Cal, create Track issues, run Flow automations and hand off to a person.

Missing something?

Tell us what to build or fix next. Every message gets read.

Suggest an idea