Release by release

Agents that plan, delegate and build

Agents can now split work into parallel workers, write a plan and run it, hand work to other agents, and build and run full websites from a description. Organizations get a proper home with a chat you can act from, rules, notices and live status, and Tools now run like a simple form that shows its progress.

Why it matters: agents finish bigger, multi-part jobs on their own, while people keep clear control of what gets sent, spent and approved.

New

• Workers and plan-then-run. A Prompt AI step can split its job into short-lived workers that run in parallel, or write a plan of up to five phases, including a "check" phase where reviewers challenge earlier results. You see the plan and its estimated cost before it runs, it stays within a credit budget, and a plan that worked can be saved as a series.
• Steps that use tools. A Prompt AI step can call the agent's own series as tools until it has an answer. Each call is recorded as its own run, with its own approvals.
• Hand-offs between agents. Org admins can let one agent pass work to another. The receiving agent does the work with its own permissions, and anything it sends or writes waits for its admins' approval by default.
• Run tree. A run's page shows everything it started, how each part ended, what each cost and the total.
• A new Tools page. Questions come one at a time, you review your answers, then run. Progress shows live step by step, a "Running now" list tracks active runs, and leaving the page doesn't stop a run.
• Sites from a description. Describe a site and get a working one, with data and sign-in included. Each site has Overview, Analytics (people vs AI assistants), a form inbox, Data, Users, Security, Deployments and Domains, plus a launch checklist and a brand kit.
• Ask your organization anything. An org-level chat that runs series and follows them, completes approvals inline and creates sites. It suggests steps for you to confirm, keeps conversations, takes voice input, and answers /commands (/run, /approvals, /pause…) instantly.
• Organization home. A daily brief, proposals you can act on in one click, questions from your agents, live activity, a status strip of counters you choose, and up to eight pinned panels.
• Rules and notices. "When this happens, do that" rules can notify people, run a series, require approval before a step that sends or writes, or block automatic runs. Default notices tell you when a run fails, an approval is decided or an AI key stops working.
• Integrations, reworked. A catalogue of apps, connections for the whole organization, one agent or just you, connection health with Reconnect, and permissions for each connection. Includes Gmail, GitHub, HubSpot, Odoo and Airtable steps, and search across your connected apps.
• Describe your agent. Draft a whole agent from a description (brief, guardrails, starter series and knowledge), review every part, then create it. Agents can be imported and exported as SKILL.md files.
• Set up an organization from its website or a short description. You get a proposed team of agents, each with a first job.
• Logic and data steps. If / Else, Switch, Merge, For each record, Pick fields and Data output, plus retry, time limits and error branches on any step. Every step gets an Output tab (Table, JSON, Fields, CSV).
• Choose your AI. Claude (Anthropic) can be an organization AI provider, and you can pick the model per organization, agent or step. Credit pricing is clearer, by model class.
• Also new: artifacts (shareable pages agents publish), the Everyday home feed, dark mode everywhere, federated search, series scheduling, private-by-default sharing with roles, and team email (SMTP).

Fixed

• Schedules run once per occurrence, never early or twice.
• Pausing agents also stops runs started by schedules and webhooks.
• An AI answer that starts with "Error" no longer stops the run, and approval now comes before a step that sends.
• If / Else passes its input on, instead of "true" or "false".
• Running out of credits during PDF reading or Python steps now stops the run cleanly.
• Writer sharing, share links and PDF export work as described.
• Under the hood: a large security review closed dozens of issues (tenant isolation, secret handling, access checks, request limits). Runs and history are faster and recover better from outages.

Smarter approvals, sturdier series, self-hosting

Approvals now work the way real teams do: one-click email approval, hand-over, group approval, escalation and expiry. Series gained if/else branching and field-level data passing, decks got a hands-on editor, and the platform can run fully self-hosted.

Why it matters: moving a process from a demo to daily use needs approvals that don't stall, workflows that make decisions, and a deployment that fits your own infrastructure.

New

• Approvals by email, with one-click Approve and Reject. In-app notifications arrive even if browser notifications are off.
• Flexible approvers. Hand an approval to a teammate, assign it to a group where the first person to act decides, escalate on a timer, or set an expiry with a default path.
• If/else branching: a Conditional step sends a run down different paths depending on the data.
• Use one field from an earlier step, e.g. {{Read order.customer_id}}, with | json for API bodies. A missing field stays visible instead of turning blank.
• Outlook add-in: send the open email to a series, which can extract the details and route them for approval.
• Faster starts: ready-made starters, a first-run checklist, and AI-suggested series that build themselves when you click one.
• Deck editor: edit text on the slide, alignment guides, drag-to-reorder, insert between slides, and a present mode.
• Agent chat keeps its history, with a Chat Logs panel.
• New steps: Fern AI (your organization's fine-tuned models), Appwrite (site data and users), and Email Output with HTML, CC and multiple recipients.
• Self-hosted deployments: an enterprise mode runs on your own infrastructure, with server health checks.
• Connect all your agents to Claude at once with one claude mcp add command.
• Smoother onboarding: the app reopens your last team, and you can invite from the agent header.

Fixed

• A failed step now stops the run, and history shows the real error and output for every step.
• Errors in AI writing, PDF export, invites and chat show clear messages, and chat lets you retry.
• Switching organizations no longer signs you out.
• Saving a series no longer wipes typed input values.
• One Save button in Settings, API keys included.
• Many dark-theme contrast fixes.
• Under the hood: a major security pass on team, site and organization permissions, keys kept on the server, seat limits, suspended subscriptions, sign-in throttling and outside-service timeouts.

Integrations hub, smoother billing

Connect GitHub, GitLab, Gmail/Google, Slack, Notion, Jira, Confluence and Trello from one Integrations page, and use GitHub, Gmail, Jira, Notion and Slack as steps in your series. This release also fixes AI Credits balances, plan limits and email delivery.

Why it matters: your series work in the tools your team already uses, and your balance and limits are always right.

New

• One-click Connect for GitHub, GitLab, Google/Gmail, Slack, Notion, Jira, Confluence and Trello.
• GitHub, Gmail, Jira, Notion and Slack steps for any series.
• A clear message with an upgrade link when you reach a plan limit.
• An improved billing page; plan changes and payments update your account correctly.
• A redesigned Agent Settings → General.
• Refreshed sign-in, verification and report emails.
• Newer Gemini models (Gemini 2.0 is retired).
• The AI Credits balance updates live.

Fixed

• The credits balance only updated on reload.
• Super admins couldn't change a user's credits.
• Platform emails often landed in spam.
• Creating an agent sometimes failed with "invalid JSON".
• Publishing a series sometimes failed.
• Sign-out in the mobile app was unreliable.

Agent memory, voice and a new design

A large release. Everyday has a new look across Studio, Stage and Solutions, and agents now have memory and voice. Series can include other series, and agents can connect to Claude and other AI assistants.

Why it matters: your agents remember your context, you can talk to them and plug them into other tools, and setting up work takes fewer steps.

New

• Agent memory: memory notes for the whole organization, one agent or yourself, used by every AI step automatically.
• Voice sessions: talk to your agent live, with a transcript.
• Series inside series: use one series as a single step in another.
• Connectors: secure links that connect your agents and published series to Claude and other AI assistants, plus a skills file and an API reference for each agent.
• New design: theme, sidebar, organization and agent dashboards, and quick switching between Studio, Stage and Solutions.
• Sites: redesigned creation, a deploy indicator, an environment-variables editor (paste a whole .env file), and request-a-site.
• Writer: a personal workspace with tags, search and grid view, and a tool picker that inserts results into your document.
• Tools and series: a security check before publishing, automatic thumbnails, and private/shared/published filters.
• Files: multi-select and ZIP download, plus real storage usage.
• Billing and usage: a plan comparison table, provider filters, the savings from memory caching, and agent availability.

Fixed

• Series crashed when you changed their tag.
• Loaders spun forever after switching organizations.
• The run screen didn't reset between runs.
• Nested series ran out of order.
• White screens now show a recovery screen instead.
• Under the hood: rate limiting, input validation, stronger encryption and faster lists.

Decks, Sites and flexible AI setup

Decks, a slide builder that can show live series results, and Sites, which creates and publishes websites from a marketplace that works with your own Appwrite setup. AI setup is more flexible: use any OpenAI-compatible provider and pick models from a visual list.

Why it matters: turn series results into presentations and live websites without leaving Everyday, on the AI provider you already pay for.

New

• Decks: slides with charts, icons, images and live series results. Tag, import, duplicate, and export to PDF.
• Magic deck and Ask me anything: AI drafts the slides, or suggests changes you preview before applying.
• Sites and the Marketplace: create and deploy sites, see their users and data, on custom or self-hosted Appwrite.
• Organization AI keys: OpenAI or any OpenAI-compatible provider, with a key test and a searchable model list.
• Prompt AI steps show their provider and model.
• Series assistant (experimental): describe a change and AI applies it step by step; type "/" on the canvas to add a step.
• Click Video Q&A: AI writes questions from a video's transcript.
• Tools: paged run history, and runs can pause for a person's input.
• My Assets: grid and list views, better thumbnails and shareable links.
• A What's new window and per-organization feature switches.

Fixed

• Uploads, thumbnails and PDF downloads are more reliable.
• JSON Splitter paths can be edited inline.
• Writer's "continue writing" now uses your organization's AI key.
• Sidebar flicker, and theme colours that only updated on refresh.
• Safari is now supported.

Foundations: series, tools and teams

The first complete version of Everyday Series. Build a visual series step by step, publish it as a Tool your team can run, and keep the results in one workspace, with chat, documents, sites and files.

Why it matters: it's the base everything else builds on — design AI work once, and the whole team reruns it.

New

• A visual series editor that runs step by step, with progress shown at each stage.
• Input steps: text, multi-field forms (text, number, choices, rich text, image, audio, video, PDF), CSV and image.
• AI steps: prompts, image and media models, and text analysis (language, sentiment, key phrases, entities, personal data).
• Logic and data: API calls, Python, JSON splitting and Markdown formatting.
• Outputs: text, Markdown, HTML, image, video, audio, email, a Writer document, or a blog post.
• Describe a task and get a first draft of the series, or start from a template.
• Tools: publish a series as a simple form, with run history.
• Writer, Sites, Assets and Tags.
• Teams (now called agents), with members, invites, roles and billing.

PDF reading fixed

Reading text from PDFs in series works again.

Fixed

• Reading text from PDFs in series works again.

Forms and templates

Series can start with a form, and creating a series from a template is easier.

New

• Series can start with a form, and creating a series from a template is easier.

Version in the sidebar

The sidebar shows which version you're using.

New

• The sidebar shows which version you're using.

Release records

Release numbering and change records switched on.

New

• Release numbering and change records switched on.

First tracked release

The first release with a version number.

New

• The first release with a version number.