Moonshot AI's open-weight frontier model is now available: strong reasoning and tool use, a 1M-token context window, and inference in the EU (France).
Agents can now use MCP servers (Model Context Protocol): connect a server, discover its tools, and approve exactly which ones the agent may use. Custom HTTP tools also got a clearer setup.
New models via Prem, a Swiss provider with an extra security layer: requests are end-to-end encrypted and processed in hardware-sealed enclaves, with inference entirely in Switzerland.
Follow-up messages can now be sent while the agent is still replying: they are queued visibly and answered together in one coherent reply once the current answer finishes.
Agents can now search the web live and answer questions about current topics with source citations. Web search can be enabled per agent and restricted to specific domains.
Documents can be uploaded directly in the chat — the agent reads them and answers questions about their content. File uploads can be enabled per deployment.
Knowledge bases are more reliable and easier to manage, with steadier crawling, instant deletes, and clearer indexing status.
A new documentation site with full-text search (⌘K), dark mode, and guides for every feature.
Teammates can be invited as Members to chat with the agents without seeing the builder dashboard.
Website and internal-chat deployments can set a backup model that takes over automatically if the primary AI provider has an outage.
Previously deactivated members can be re-added or reactivated in one click — seat limits are respected.
Every conversation across website, phone and internal chat in one place, with channel and source filters.
A Server Location tier (Swiss, European or global) limits which AI models can run, and a per-deployment Data Assessment shows which countries the data touches.
Organizations can choose whether admins see internal-chat content or only metadata; each chat locks in the policy active when it started.
A new cost calculator and clearer credit pricing across the app, with per-model rates admins can review.
Knowledge bases now show page and upload-storage usage with clear per-plan limits.
AI generation, transcription, phone-number and chat errors now show a clear reason instead of failing silently.
The dashboard chat can be opened directly on a specific agent or model with ?agent= and ?model= links.
Plans are now Free, Lite, Standard, Pro and Ultra, with clearer limits.
The agent page is reorganized into clear Configure, Deploy, Contacts, Forms and Usage tabs.
A refreshed landing page and a new Trust / Security page.
Launched Web Page Roast, a free public tool that gives a quick conversion analysis of any web page.
A form builder with built-in templates (contact, support ticket, booking), plus a submissions tab with CSV export.
Internal chat gained continuous conversations plus richer configuration — avatar, welcome message, and a show-tool-use toggle.
Thumbs-up/down feedback on agent replies and the ability to flag conversations for review.
A public models page showing each model's credit pricing, speed and intelligence.
A realtime voice test panel with a live mic visualizer, plus an opt-in call-recording toggle with playback.
A People view shows who each agent has talked to, and with Customer Memory the agent recalls preferences and context across conversations.
A one-sentence description of the agent generates a tailored system prompt — with ready-made templates and voice dictation.
A conversations view with full transcripts, channel and phone-number context, and tool-call details.
Agents can be deployed to a real phone number — with voice selection, turn-detection and call settings, and numbers provisioned from the organization's pool.
Agents can search content in hybrid, semantic or keyword mode — with clearer crawl status and a built-in search tester.
Sign-in with passkeys (WebAuthn) and two-factor authentication are now available, alongside a new Trust Center.
BOTTS.AI