Day AI Release Notes

How To Build The ULTIMATE AI Company Brain (Obsidian + Claude Code)

Your second brain is single-player. This is the multiplayer version.

This one was fun to make. I spend the whole film playing Althea, Director of RevOps at RMBL, a fictional company that makes roller coaster management software and is trying to get a new water park product into every deal. Althea starts where most ops people actually start: a folder full of call notes, a CRM export, somebody's pricing scratchpad. One prompt into Claude Code and that mess becomes a clean, interlinked Obsidian vault, the Karpathy LLM-wiki pattern, with a graph view you can wander around all day. You can build that part in an afternoon, and I show you how.

Then the leap. I point Claude Code at GTM Brain, our public repo, run /start, and it connects the vault to Day.ai. It reads the initiative I care about, drives attach rates on the new product, audits the whole sales team against it, and catches that Mason, the new rep, has no agent at all. A few approvals later the rollout ships: Mason gets an agent built from what's working across the team, everyone's call prep quietly learns when to bring up the new product, and every change lands in a change set I can roll back. One shared brain, every agent owned by a human, deployed from my terminal.

What Shipped

Your Agent Learns Your Voice Automatically

Day AI now builds a personal Email Writing skill for every user — automatically. Once your Gmail finishes syncing, your agent reads dozens of your real sent emails, studies how you actually write — tone, length, greetings, sign-offs — and rewrites its own drafting instructions in your voice. You get a /email skill without setting anything up. No style-guide homework, no prompt engineering.

The skill also knows your email templates. It lists each one by name with guidance on when to use it, and pulls the current version at draft time — so when you edit a template, every future draft picks up the change. And because it's a normal skill, you can refine it in plain language: tell your agent "make my emails shorter" or "stop using exclamation points" and the skill updates.

Also shipped

• Skills can include other skills: A skill can now reference up to five other skills belonging to the same agent, folding their instructions in at run time. Keep your "Email Voice" guidance in one skill, include it from "End of Day Email" and "Follow-up Drafts", and refining the voice once improves everything that uses it.
• Widescreen mode: Press Ctrl/⌘+W to let chat, Pages, and Home stretch to the full width of your window instead of a centered column. Toggle it back the same way.
• Markdown uploads in chat: You can now attach .md files in chat alongside the other supported file types, and your agent reads them directly.
• Copy and export Pages: Pages get Copy page and Download as Markdown actions in the panel menu — pasting into Docs or Notion keeps the formatting, and plain-text editors get clean markdown. Individual tables get their own Copy action too, preserving headers, colors, and column widths when pasted into another Page.
• Better folder browsing: Folder cards in Pages now show relative last-updated times, wrap longer titles instead of truncating them, and can be sorted by folder order, name, or last updated — and inline folder expansion now actually shows the folder's contents.
• Decision tree charts: Ask your agent to diagram branching logic — lead qualification flows, routing rules, playbooks — and it renders a proper decision tree with labeled branches, exportable and embeddable in Pages like every other chart.
• New connectors — Crossbeam, Sumble, Common Room, and Guru: Four new MCP connectors join the roster. Connect them from Workspace Integrations with per-user OAuth and your agent can work with partner ecosystem data (Crossbeam), technographic data (Sumble), community signals (Common Room), and verified company knowledge (Guru) directly.
• Top-N dashboard tables: Table widgets accept an optional row limit applied after sorting, so you can build an honest "Top 20 deals by amount" table — with the header showing "Top 20 of 312" so nobody mistakes it for the full list.
• MCP tools from Slack: Your agent can now use your connected MCP tools when you talk to it in Slack — always-allowed tools run silently in 1:1 DMs, and tools that need permission post an approval card right in the Slack thread with Approve/Deny buttons. In shared channels, every connector call asks first and the card discloses where the result will be posted. [needs context: confirm rollout scope — currently enabled for select workspaces]

Know someone else who should check out Day AI? Shoot them this link:
https://day.ai/demo

Original source

No more one by one merging required. Tell your agent to clean up duplicates by a shared field — email for contacts, domain for organizations. It previews the plan, you confirm, and merges run in the background, skipping any group it can't merge without losing data.

Also shipped:

• Page folders — You can now group your pages into private and shared folders, allowing for easy organization of proposals, reports, knowledge documentation and more.
• Page table polish — Tables are left-aligned and fill the page by default, with a new alignment and sizing menu, per-column and per-cell text alignment, and a richer formatting bubble (bold, italic, underline, links, dark-mode-safe text color).
• Page Exports — Share links and PDF exports now render tables exactly like the editor.
• Email citations — When your agent states a fact drawn from an email, it drops a numbered citation chip next to the claim; click it to open the exact email in the sidebar.
• Big attachments, finally readable — Attach images up to 30MB and PDFs up to 50MB in chat. Images resize instead of erroring, long PDFs read page by page.
• Opportunity share links — Ask your agent for a share link to an opportunity and you'll get one that opens the deal with its activity view alongside.
• Stage-change history — Your agent can read the change history of one property across many records at once — ask when each open deal last changed stage and get a source-attributed answer.
• Smarter matching for government domains — dmv.ca.gov and sanbruno.ca.gov are now distinct organizations instead of collapsing into ca.gov, so activity lands on the right record.

Original source

Answering questions is the tip of the iceberg

Answering questions is the tip of the iceberg when it comes to what a Day AI agent can do. The real leverage is treating your agent like an employee with a role and giving it skills: recurring jobs it owns without you asking each time.

A skill doesn't wait in a chat. It runs on a schedule or in response to something happening in your workspace, does the work, and tells you what it found or did.

They already know.

Every skill runs on your agent's memory of your customers — built continuously from the emails, meetings, and Slack conversations you've connected. That memory keeps accumulating whether or not you're actively using it.

Set it and forget it.

Set a skill up once and it just keeps happening. Nobody has to remember to do the work, because it's already done by the time you'd have gotten to it.

Setting a skill up is a conversation, not a project. Describe the outcome, your agent builds the skill. No configuration screens to learn first.

"Create a skill that sends me a weekly pipeline summary every Monday at 9am."

"Create a skill that drafts follow-up emails after my meetings."

"Create a skill that flags bug reports from Slack."

You don't have to build a skill from scratch. Day AI ships ready-made skills — Actions, Opportunity Automation, Call Coaching, Deal Coaching, End-of-Day Follow-Up Drafts, Hot Accounts — that you can add to your agent in minutes and adjust from there.

A working skill fades into the background. You'll know it's working when you stop doing the manual version of the task and only think about it when the notification lands. Still doing it by hand "just in case"? The prompt needs a tweak — tell your agent what's missing.

Fixing a skill works the same way. Your agent helps shape the prompt, pick a trigger, and set up notifications up front — and if something's off once it's running, you tell your agent what should change instead of hunting through settings.

For the mechanics (trigger types, data privacy settings, the Skills Library, troubleshooting a skill that isn't firing) see our Skills help doc.

Connectors: where the payoff compounds.

Skills are the most powerful when they can use your other tools. Without connectors, a skill can watch what's happening and tell you about it. With them, it can take action. This is where skills stop saving you a glance and start saving you the work.

• Customer mentions a bug on a call? Your agent opens the Linear ticket.
• Deal moves forward? It updates the stage in HubSpot.
• Meeting wraps? The follow-up is drafted and the recap is in Slack before you've closed the tab.

Bring your customer memory anywhere: Day AI via MCP

Day AI's customer memory is retrievable via MCP — the open protocol that's quickly becoming the standard way AI tools connect to live data, and that Claude, among others, has built native support for.

That means your workspace's memory isn't trapped inside the Day AI chat window. Connect it once, and you can open Claude Desktop or Claude Code and ask "what's the latest on Acme Corp?" and get an answer sourced from the same real-time customer graph your Day AI agent uses, without ever opening Day AI.

No exporting a CRM report. No copy-pasting a transcript into a prompt. No re-explaining who the account is or where things left off. The context travels where you're already working.

Most CRMs make you come to them. Day AI goes where you are. Set the connection up once, in a couple of minutes, and it's just there from then on.

Original source

The Agent Origin Story

How a morning briefing nobody was supposed to reply to changed the course of the company, and what it means for the future of work.

Every company has the moment where everything clicked. This is ours.

It started with a morning briefing email that nobody was supposed to reply to. Then everybody replied. And one person in particular kept replying, even after Christopher told him to stop.

What looked like a support problem turned out to be the insight that built the entire agent architecture: when an agent reaches out to you, you respond to it like a human. You coach it. You give it direction. You expect it to follow that direction, and follow it well.

That expectation, that agents belong to people, that they have job descriptions, that they get better over time through real feedback, is the foundation everything at Day AI is built on. From morning briefings to skills to the GTM Brain. It all started here.

Email joins the timeline

Email is where most of your customer relationships actually happen. Day.ai already captures your email automatically. Now we're putting all of it to work in two big ways.

See every email in the timeline

Open any person or organization and the timeline now shows your email history right alongside meetings and everything else — every message at the moment it was actually sent, with threads grouped so your timeline stays readable. At a glance you'll see what was sent and received, and clicking into an email opens the full conversation: every participant, complete message bodies with inline images, and attachments you can preview and download.

Ask your Agent to find any email

Your Agent can now search across all emails that you have access to, lightning fast. Ask things like:

• "Find the pricing emails we exchanged with Acme last quarter"
• "What did Sarah say in her last email about the renewal?"
• "Pull up every email with anyone at bigco.com from the last two weeks"

Search by who sent or received it, by company, by timeframe, by subject or content — and get the actual messages back, not just a summary. It's also significantly faster: we rebuilt email search on our new search infrastructure from the ground up.

Private stays private

Your email sharing rules apply everywhere, always. Messages you haven't shared with your workspace are clearly labeled Private and visible only to the people on the email — and anything covered by your exclusion rules (a domain or a specific address) kept private to you and your Agent. You can review or update your sharing rules anytime, or ask your Agent to update for you.

Also shipped

• Onboarding without Google — You can now finish onboarding and use Day without connecting a Google account. A confirmation explains what you'll miss, and you can connect later from Settings.
• Intercom & Amplitude connectors — Connect them as MCP connectors and let your agent work with your support conversations and product analytics directly. Per-user OAuth from Workspace Integrations, like the other connectors.
• CSV exports show names, not IDs — Multi-select picklist columns were exporting raw option IDs for cells with 2+ selections; exports now show the option names you see in the table.
• Page share notifications — When someone shares a page with you directly, you get a bell notification. Mentions also check real page access first.

Know someone else who should check out Day AI? Shoot them this link:
https://day.ai/demo

Original source

Quick answer: A context graph is a way of storing information about your customers as a network of connections, people linked to companies, linked to deals, linked to every conversation anyone has ever had about them, instead of as rows in separate tables. The point isn't the shape. The point is that when something changes anywhere in that network, every agent looking at it sees the update immediately, and can follow a connection to any related fact without a person having to go find it first.

That's the short version. The longer version is worth understanding, because "graph" gets thrown around loosely, and the actual property that matters is easy to miss if you only hear the word.

The problem with tables

The problem with tables

A traditional CRM, and most business software generally, stores information in tables. A contacts table. A companies table. A deals table. Each row is a record, each column is a field, and if you want to know how a person relates to a company, or a deal relates to a conversation, you're relying on a foreign key, an ID number sitting in one table that happens to point at a row in another.

This works, but it's brittle in a specific way: the relationships are implicit. The database knows a contact_id equals 4471 and a deal has a contact_id field set to 4471, but it doesn't inherently know or care that this means "this person is the champion on this deal, and three months ago they raised a concern about pricing that never got resolved." That meaning lives in a person's head, or in a note somewhere, disconnected from the structural fact that connects them.

What a graph does instead

What a graph does instead

A context graph makes the relationships themselves the primary thing being stored, not an afterthought bolted on with foreign keys. A person links to a company. That company links to a deal. The deal links to every conversation anyone on the team has ever had about it. Each of those links, not just each record, carries meaning: this person works at this company, this conversation happened on this date and touched on this deal, this commitment was made and hasn't been followed up on.

Practically, this means an agent (or a person) can start at any point, a person, a company, a deal, and walk outward to everything connected to it, the way you'd follow a thread of "oh, and that reminds me" in a real conversation, instead of running a separate query against a separate table for every fact you need.

Here's a concrete version. Say a contact at Acme Corp asks a question about a feature. A relational system can tell you who they are and maybe what deal they're on. A context graph can tell you: this person works at Acme, Acme purchased the Enterprise package eight months ago, that purchase generated two support tickets about integration issues, one of those tickets is still open, and this same concern was raised in a call last quarter that flagged renewal risk. That's not five separate lookups. It's one connected picture, because the graph already holds the connections, not just the facts.

Why this matters specifically for agents

A person can tolerate a database that only answers exact questions, because a person can go dig around when something doesn't fit a field. An AI agent can't dig around the way a person does. It needs to be able to query in something close to natural language and get back not just a data point, but the surrounding context, and then be able to follow that thread to nearby, related information on its own, quickly enough that a real-time conversation or workflow doesn't stall waiting on it.

That's the actual design requirement a context graph is built to satisfy: query in plain language, get back a summarized answer along with the underlying facts, and be able to walk to connected nodes to assemble whatever context window is needed for the task at hand, fast. A flat table, or a folder of documents with a search function bolted on, can approximate small pieces of this, but it doesn't hold the relationships as structure the way a graph does, which means every question that spans more than one type of record turns into manual reconstruction work.

The part a homemade version tends to miss

If you've tried building something like this yourself, notes and call transcripts dropped into a folder with an AI model pointed at them, you've probably gotten a version of the retrieval benefit already. That part is real, and it's a reasonable place to start.

Where it tends to break is on three properties that don't show up until a second person, or real customer data, gets involved:

Provenance and lineage.

Not just what the system believes is true, but where that belief came from, when it was established, and whether a human ever overrode it. "We think this deal is at risk" is a much more useful and trustworthy statement when you can also see it was flagged after a call three weeks ago, was correct twice before, and a manager confirmed it last time. A flat document store doesn't track any of that by default.

Permissions, enforced at the level of the graph itself.

Some information should be visible to everyone on a deal. Some shouldn't. A homemade system built for one person doesn't have this problem, because there's only one person's access to think about. The moment a second person, let alone a whole team, needs to use it, every question the graph answers has a second question hiding inside it: is this specific person allowed to know this specific thing? Enforcing that at the level of the underlying graph, rather than bolting permission checks onto every application built on top of it, is a structurally different, and harder, problem than most DIY versions ever attempt.

Speed at the moment of retrieval.

Walking from a person to a company to a deal to a conversation and back, in something close to real time, while also checking permissions on every hop, is a genuine engineering problem, not a configuration setting. It's the specific place a homemade index, built to answer one kind of question well, tends to fall over once the questions get varied and the data gets large.

Context graph vs. customer memory: how the two terms relate

These two terms get used close enough together that it's worth being precise. Customer memory is the broader idea: a continuously updated record of everything that's happened with a customer, captured first and structured whenever needed, instead of structured in advance. A context graph is the specific technical shape that record takes, the nodes and the edges, the relationships between people, companies, deals, and conversations, that makes that memory something an agent can actually query and walk through quickly. Customer memory is the what and the why. A context graph is the how.

Frequently asked questions

Is a context graph the same thing as a knowledge graph?

They're closely related concepts. A knowledge graph typically models general facts and entities. A context graph, in the customer relationship sense, is a knowledge graph specifically structured around people, companies, deals, and the conversations connecting them, with the retrieval speed and permission model built for agents to query it live.

Do I need a graph database to build a context graph?

Not necessarily as a strict requirement, but you need something that treats relationships as first-class structure rather than as foreign keys bolted onto separate tables, and that can traverse those relationships quickly at query time. Whether that's implemented on top of a graph database specifically or another architecture that achieves the same properties is an implementation detail; the properties (queryable relationships, provenance, permissions, speed) are what actually matter.

Can I build a context graph myself with notes and an AI model?

You can build something with real value for one person this way, and it's a completely reasonable place to start. It tends to hit real limits once a second person needs access (the permissions problem), once you need to know why the system believes something and whether a human corrected it (the provenance problem), or once the volume of data makes fast traversal a genuine engineering challenge rather than something a script handles.

Why does this matter more now than it did five years ago?

Because the thing querying the data changed. A person clicking through a dashboard can tolerate flat tables and manual lookups. An AI agent that's supposed to act on your business needs to retrieve connected context fast, in plain language, without a person assembling it by hand first, and that's a different architectural bar than most systems were ever built to clear.

Day AI's Context Graph is the foundation every agent in the product runs on, continuously updated from every call, email, and thread, with provenance, lineage, and permission-aware access built in from the start.

See how it works →

Original source

Software Can Read Now

Every call you take, every email you send, every decision your team makes — it's all being captured. The question is whether any of it is doing anything.

CRM was built to store that data. It was never built to act on it. What's happening now is that the monolithic suite is breaking apart into categories that are each becoming real products in their own right, and the first company to reassemble them into one system, built from the ground up for the agentic era, becomes the successor.

That's Day AI. Every conversation your company has automatically closes the loop. A bug surfaces on a call? Engineering gets a PR to review. A prospect mentions how they found you? An agent starts pulling that thread. A deal goes quiet? A re-engagement is drafted and waiting for your approval. No prompt required. No one has to remember to check.

Mention a teammate in a Page — and give them access in the same click

Pages just became a place to work together, not just write.

You've always been able to capture knowledge in Pages. Now you can bring the right people into it. Today we're introducing @-mentions in Pages — and we built them to do something most tools make surprisingly hard.

The old way:

To loop a teammate into a document, you'd stop writing, open a separate share dialog, guess at the right permission level, save, and then message them somewhere else to let them know. Four steps, three of them outside the doc.

The new way:

Type @ and pick anyone in your workspace. Hover the mention and you'll instantly see whether they can actually access the page — and if they can't, grant or adjust their access right there: Viewer, Collaborator, Editor, or Full access. The moment you mention them, they get a notification in the new notification bell with a link straight to the page. Mention, grant access, and notify — all in one motion, without ever leaving the page.

You can also type @ to link to other Pages in your workspace, so related docs stay connected.

Also Shipped

• Email Privacy Recommendations — Each time we have a new email sharing exclusion recommendation for you, you’ll be notified and clicking the notification will take you to email exclusion settings.
• Attio connector — Connect Attio as an MCP connector and have your agent work with your Attio CRM data directly. Authenticate per-user from Connectors, same OAuth flow as the other MCP connectors.
• Slash command menu on Pages — Hit / to open the menu and insert a new text or image block, table or more without using your mouse.
• Stage badges on related opportunities — Opportunity in the Related tab now show a stage chip alongside the pipeline name, so you can read pipeline status without opening each one.
• Export to PDF — Hit Save as PDF in the three dot menu of a Page for a clean print-ready export. It works on public shared pages too, no login required.
• Sidebar remembers your filter — The detail-view sidebar reopens to the property filter tab you last used instead of resetting to "All".
• Annual commitment visibility — The Billing page now shows per-tier commitment usage, and add/remove-agent dialogs spell out when you're filling a paid slot vs. leaving one unfilled.

Original source

This week Laura and I sat down to answer a question that's been nagging at every GTM engineer I know: you've migrated your tinkering into Claude Code. Doing what, exactly? Chapter one of GTM engineering (which Clay largely defined) was hands-on data enrichment: powerful, creative, project by project. What chapter two actually looks like is a different thing entirely.

We rebuilt table editing on Pages from the ground up — tables now work the way you'd expect from the best modern editors: flexible, fast, and fully editable by you or your Agent.

What it changes:

Shape your columns instantly.

Drag any column border to resize it. Double-click a border and the column auto-fits to its content. Want the table to fill the page? One click on "Fit to width" and every column balances itself.

Restructure without friction.

Hover over any row or column and a grip appears. Click it for a full action menu — insert, duplicate, clear contents, or delete. Toggle the first row or column into a header. Need five more rows? Click the + bar below the table to add one, or drag it to add as many as you need in one motion.

Make tables scannable with color.

Apply background colors to any row or column — nine colors, all tuned for both light and dark mode. Status columns, priority rows, highlighted totals: your tables can finally communicate at a glance.

Move tables like blocks.

Grab the handle beside any table and drag it anywhere on the page. A drop line shows you exactly where it will land.

Pages can now include images right alongside your written content.

You can paste, drag and drop, or use the insert menu to add screenshots, diagrams, mocks, inspiration, or visual notes directly into a Page. Once added, images save with the page and can be resized or aligned left, center, or right.

What it changes:

A lot of work is easier to understand visually. Instead of keeping screenshots in Slack threads, docs, or separate tools, you can now keep the image and the surrounding context together in
Day.ai
. This makes Pages better for recaps, plans, research, customer notes, product feedback, and any work where a picture carries important context.

• Clay Connector — Add Clay as an MCP connector and have your agent work with your Clay data directly.
• Connector guardrails in skills — You can't add a connector you haven't actually connected; unconnected providers show an inline Connect button so the skill won't silently fail at runtime.
• Salesforce Classic Notes import — Bulk-import your Salesforce Classic Notes from a CSV; Day creates any missing orgs from the note domains and imports each as an org-linked note.
• Former owners stay visible on Opportunities — Opportunities owned by someone who left the workspace now show that owner in a single clean picker.
• One reply from your Agent when you chat by email — Instead of 3–4 back-to-back messages per question, the agent works in the background and sends one complete reply at the end.

Know someone else who should check out Day AI? Shoot them this link: https://day.ai/demo

Original source

Five levels. One decision.

Anthropic. OpenAI. The whole ecosystem is handing you tools and telling you the door is open. And they're not wrong, you can start it. Claude with MCP, a few integrations, some prompt engineering. You can absolutely get something running.

That's the trap.

Because the gap between what you can start and what you actually need is enormous, and it doesn't reveal itself all at once. It reveals itself slowly, in the form of data you can't access, permissions you can't navigate, product decisions you don't know how to make, and customer-facing quality that's just a little off in ways that compound quietly over time. By the time you feel it, you're already behind and underwater.

There's a reason software engineering teams didn't just build Cursor or Claude Code themselves. Not because they couldn't start, they could. Because getting to best-in-class requires depth, cycles, feedback, and time that compounds in the right direction. DIY gets you something. It doesn't get you there.

Five levels. One decision. Everything on that ladder exists in Day AI today and teams that are running it aren't thinking about whether they can build this. They're thinking about what to hand off next.

Also Shipped

• Create organizations without Domains.
  Previously a domain was required to save an Organization in Day AI, but sometimes you work with organizations without websites. Now you can save those organizations

• Pipeline: filter by multiple owners.
  The board Owner filter now supports "is any of," so you can filter to a group of reps at once instead of one at a time.

• Apollo MCP Connector.
  You can now search prospects, enrich people and companies, manage contacts, and work with Apollo sequences directly in Day AI in chat or Agent skills.

Know someone else who should check out Day AI? Shoot them this link:
https://day.ai/demo

Original source

Customer memory isn't just a record of what people said — it's knowing who they are. The company, the deal, the bug three people hit, the Slack thread, the moment in the meeting recording where it happened. All of it structured, all of it live in Linear, none of it entered manually.

Linear's customer requests feature is one of the most underused tools in product development, not because it's bad, but because most teams can't keep it fed. Day AI solves that. Every conversation, feature request, and support issue flows in automatically, tied back to the source, linked to the people behind it. The customer has a name, a face, a company, a reason they care.

When your whole team has that context, everyone builds like a forward deployed engineer. You're not triaging abstract tickets — you're solving real problems for real people you actually know. That changes how you prioritize. It changes how you build. It makes the work more human and, honestly, more fun.

That's not a vision. That's what’s featured here. And we set it up in about 10 minutes.

People Without Email

You can now add People to Day AI before you have their email address.

Not all relationships start with email – sometimes you met someone at a conference and only got their phone number, or you reached out to them on LinkedIn and only have their profile URL. These are people you want to store and track in customer memory, and now you can.

What it changes: no more placeholder addresses, no more workarounds, no more "I'll add them later" that never happens. The person is the record. The email is just one more field.

Stay tuned for adding Organizations without a domain, coming soon!

Also shipped

People Without Email

Now you can add People to Day AI without having their Email Address–whether you have a linkedin profile, a phone number or just a name, creating a more complete picture of your relationship graph.

Email search now sees the whole history of a merged contact or company

When you merge two contacts or two organizations in Day.ai, you're telling us "these are the same person/company — treat them as one." Email search now honors that.

Email Signatures

You can import your signature from Gmail to be appended to all emails sent via Day AI.

Meeting Recording ↔ Opportunity Relationship Management

You can now link meeting recordings directly to opportunities and manage those relationships — not just view them.

Workspace Member Removal

When removing a workspace member, you now get a confirmation step that also handles deleting agents they owned, so you don’t need to take two separate steps.

Original source

Skills

Skills turn your agent from a general assistant into something that can run specific workflows on your behalf. Each skill gives the agent focused instructions for a particular job.

A skill usually includes:

• Prompt: what the agent should do
• Trigger: when the skill should run
• Connectors: what the skill can use when it runs
• Notifications: where the skill can send results, when configured

For example, one agent might have a skill for meeting prep and another for drafting follow-up emails. Another agent might have a skill for sending bugs to Linear or logging feature requests.

Your agent can help you create and refine skills. You can describe what you want in plain language, and your agent can help turn that into a clear prompt and configuration.

Skills have three trigger types:

Manual

Runs when you type a /slash-command of your choosing in a conversation or ask your agent to activate the skill. Nothing happens automatically.

Event

Runs when something specific happens in your workspace, such as:

• Slack activity summary
• Email summary
• Meeting summary
• Note added
• Zapier integration
• Opportunity import
• Organization import
• Person import
• Opportunity created

Schedule

Runs on a recurring schedule, such as daily, weekdays, weekly, or every 4 hours depending on the agent's plan.

Setting Up A Skill On An Agent

Each skill belongs to an agent. You can create and manage agent skills from the agent's Skills page or by talking to the agent.

Step 1: Start A Conversation

Open a chat with your agent and describe what you want.

Examples:

• "Create a skill that sends me a weekly pipeline summary every Monday at 9am."
• "Create a skill that drafts follow-up emails after my meetings."
• "Create a skill that flags bug reports from Slack."

Your agent can help write the prompt, choose a trigger, and configure notifications. These steps are also outlined explicitly below.

Step 2: Name And Description

Name

The display name for the skill. Keep it short and descriptive, such as "Morning Briefing" or "Meeting Follow-Ups."

Description

A brief summary of what the skill does. Agents also use this to decide when to reach for the skill.

Tags

Tags are used to organize and filter skills in the Skills Library.

Step 3: Prompt

The prompt is the set of instructions your agent follows when the skill runs. A good prompt should include:

• When not to act: "If the meeting had no external attendees, stop."
• What to look for: "Find action items I personally committed to."
• What to do: "Create an action, draft an email, update a CRM field, or send a notification."
• What the output should look like: "Action titles should be under 10 words."
• When to notify: "Only notify me when something was created, updated, or needs review."

Step 4: Slash Command

The slash command lets you run the skill manually in chat, such as /morning-briefing.

Slash commands use lowercase letters, numbers, and hyphens.

Step 5: Trigger

Choose how the skill runs:

• Manual: runs only when invoked
• Event: runs when selected events happen
• Schedule: runs on a recurring cadence

Event and scheduled skills count toward automated skill limits. Manual-only skills do not.

Step 6: Data Privacy

Agent skills have a Data Privacy setting:

Workspace: The skill can search workspace-shared data.

User: The skill can access the agent owner's private context as well as workspace data.

Use Workspace for skills that only need shared CRM data. Use User only when the skill needs the agent owner's private context.

Step 7: Notifications

Scheduled and event-triggered skills can be configured to send notifications through:

• Email
• Slack DM
• Slack channel

Slack DMs require Slack to be connected in Day.ai Notification preferences. Slack channel notifications require the Day.ai Slack app to be in the target channel, which you can add by typing /dayai into the channel.

Step 8: Connectors

Agent skills can use connectors when they run automatically. Connectors let skills interact with tools like Linear, Notion, and other external systems.

Connector access is configured on the agent skill after the connector has been set up for the user. the User must have a healthy Connector status.

Step 9: Refine Over Time

You can update a skill at any time by talking to your agent.

Examples:

• "Update Morning Briefing to include Slack highlights."
• "Change pipeline review to run weekly instead of daily."
• "Don't create actions for internal-only Slack threads."

When you give feedback about a skill's behavior, your agent can update the skill prompt to reflect that feedback.

Skills Library

The Skills Library is the shared workspace catalog for reusable skills. You can find it from:

• Agents ۱۲ Skills Library
• An agent's Skills page ۱۲ Add skill ۱۲ Browse skills

The library has two main areas:

Custom: Skills created by your workspace.

Day AI: Official Day AI templates maintained by Day AI.

You can search the library and filter custom skills by tags.

Skills Library: Template vs Managed

Skills in the Skills Library have a deployment type, which determines how they behave once deployed.

Template Skills

A template skill is a reusable starting point.

When a template skill is added to an agent, that agent gets an independent copy. The copy can be edited without changing the original template or any other copies of the skill.

Use template when:

• Each user may need to customize the prompt
• The skill uses connectors
• You want the skill to be a flexible starting point

Any paid user can create template skills. The template creator or a workspace admin can edit or delete them.

Managed Skills

A managed skill is centrally controlled by workspace admins.

When a managed skill is deployed to agents, future edits to the source skill stay synced to managed agent copies. Users cannot edit managed copies from the agent profile.

Use managed when:

• Everyone should use the same version
• Admins should control updates
• The workflow should stay consistent across agents

Only workspace admins can create, edit, deploy, or delete managed skills.

Data Privacy In The Skills Library

Skills Library sourced skills run with workspace data only. Managed and template source skills cannot access an individual agent owner's private data from the library editor.

When a template skill is added to an agent, the agent's copy can be configured separately, therefore the Data Privacy setting can be configured to access user data.

Connectors In The Skills Library

Skills in the Skills Library do not preconfigure connector access.

If you copy a connector-backed agent skill into the Skills Library, it must be saved as a Template. Connectors are not copied into the library source, so each user adds the needed connectors after deploying the template to their agent.

Managed skills cannot be created from connector-backed skill copies.

Skill Tags

Workspace admins can create and manage tags for Skills Library items. Tags help organize custom skills and make them easier to find.

Tags can be used to filter the Custom tab in the Skills Library.

Deleting a tag removes it from skills that currently use it.

Adding A Skill To The Skills Library

You can add an existing agent skill to the Skills Library from the agent's Skills page.

Go to the skill, open the three dot menu in the right corner, and choose Add to skill library.

If the skill has unsaved changes, the library copy uses the last saved version. The library copy also uses workspace-level data access.

You can also create a new library skill directly from Agents ۱۲ Skills Library ۱۲ New Skill. The editor includes a Create with chat option so your agent can help write the skill.

Deploying A Skill From The Library To An Agent

From an agent's Skills page:

1. Click Add skill
2. Choose Browse skills
3. Select a skill from the Skills Library
4. Click Add to agent

For Template skills, adding the skill creates an editable copy on the agent.

For Managed skills, workspace admins can deploy the managed skill to agents. Non-admins can browse managed skills, but cannot deploy them.

Day AI Skills

Day AI provides official templates in the Day AI tab of the Skills Library. These templates are read-only and maintained by Day AI.

Current Day AI templates include:

• Actions (starts in learning mode)
• Opportunity Automation (starts in learning mode)
• Call Coaching
• Deal Coaching
• End-of-Day Follow-Up Email Drafts
• Hot Accounts
• Skill Finder

A few important details:

• Actions is available on Turbo, Professional, and Executive. It starts enabled when added.
• Opportunity Automation is available on Professional and Executive. It starts disabled and in learning mode.
• Day AI templates count toward automated skill limits when they use event or schedule triggers.
• Adding a Day AI template creates an editable copy on the agent.

Checking Skill History

Each skill keeps a log of recent runs. You can see whether a run completed, failed, queued, or is still running, and open the thread for more detail.

Go to: Agent ۱۲ Skills ۱۲ [skill name] ۱۲ History

This is the fastest way to confirm whether a skill ran and what it did.

Automated Skill Limits

Event-triggered and scheduled skills count toward a per-agent automated skill limit. Manual-only skills do not count.

• Turbo: 2 automated slots
• Professional: 5 automated slots
• Executive: 10 automated slots

Scheduled skill frequency also depends on the agent's plan:

• Turbo and Professional support daily, weekdays, and weekly schedules
• Executive also supports every 4 hours

Scheduled skills are evaluated on a 15-minute processing window.

Troubleshooting

The skill is not running automatically

Check that:

• The skill is enabled
• The agent has automated skill slots available
• The trigger matches the event you expect
• The skill belongs to the right agent owner

For example, meeting-summary skills run for meetings owned by that agent's owner, not every meeting in the workspace.

Opportunity Automation is not doing anything

Opportunity Automation starts disabled when added. Go to the agent's Skills page, open Opportunity Automation, and turn it on.

It also starts in learning mode. In learning mode, it analyzes what it would do and sends a notification for review instead of creating or moving opportunities automatically.

The skill ran but did not update the right records

Check the skill's Data Privacy setting.

If it is set to Workspace, it can only access workspace-shared data. If the skill needs private user context, switch it to User on the agent skill.

A scheduled skill did not fire exactly when expected

Scheduled skills are evaluated on a 15-minute processing window. Also check that the skill's timezone matches the intended local time.

You hit the automated skill limit

Manual skills do not count toward the limit, but event-triggered and scheduled skills do, including Day AI templates.

To free up a slot, disable or remove another automated skill, create another agent, or upgrade the agent's plan.

A library skill is not reflecting source changes

Check whether the skill is Managed or Template.

Managed skill copies stay synced with the source. Template skills create independent copies, so future template edits do not update skills that were already added to agents.

A connector-backed skill cannot be saved as Managed

Connector-backed skill copies must be saved as Template skills. Each user adds the needed connectors after deploying the template to their agent.

Questions?

Reach out to [email protected] or start a conversation with your agent and ask for help with skills.

Original source

Skill Library & Admin Controls

Day AI's Skill Library is open for business — and it puts admins in the driver's seat.

Skills like Opportunity Automation and Actions are no longer auto-provisioned onto every agent. They live in a central library where admins and reps can browse, add, and configure them intentionally. Day AI ships default skill templates that are non-editable in the library — but the moment you add one to an agent, it becomes a fully customizable, editable skill.

The bigger unlock: Managed Skills. Admins can now create a skill, lock down the logic, and deploy it across every agent in the workspace. Reps get the behavior; they don't get the edit button. When the admin updates the source, every managed copy updates with it.

What it changes:

For admins.

Define how opportunity creation, stage movement, or action generation works — once — and enforce it across the team. No more discovering a rep rewrote your pipeline logic.

For reps.

Skills just show up on your agent, already configured. You don't have to set anything up. You can still add your own skills from the library or build custom ones alongside the managed ones.

For agents.

Cleaner separation between workspace-level automation (managed by admins) and personal automation (owned by the rep). Less conflict, more predictable behavior.

Also new: An Automated Skill Slot UI that shows exactly how many automated skill slots each agent tier gets — so teams can plan their automation budget before they hit a wall.

Merge Duplicate Records

Duplicate contacts are the slow poison of a CRM. Now you can kill them from chat.

Tell your agent "Merge [email protected] and [email protected]" and Day pulls every field from both records side by side — name, title, company, custom fields, picklists, everything. Pick which value wins per row, or use one button to take all of side A or side B. One confirm and the merge runs in the background. All history, notes, emails, and relationships from the losing record fold onto the winner. No data thrown away.

Org merging is live too — same flow, same chat interface, ungated for every workspace.

Also shipped

Last Contacted At:

We added a new standard property for Last Contacted At to Organizations, People and Opportunities that is deterministically populated from Email, Slack, and Recordings – leaving no room for Agent (mis)interpretation.

Email Attachments:

Finally you can add attachments to the emails you write and send from Day AI, the same way you would in Gmail.

Kondo Connector:

For teams using Kondo to manage their LinkedIn inbox, you can now add Kondo as a Connector and use it to have your Agent draft messages, or pull LinkedIn message data into Day AI

Agent-over-email improvements:

the agent works in the background and sends one clean summary instead of a stream of intermediate thinking messages.

More automated skills!

Every Agent tier just got additional slots to run Skills on a schedule or trigger. Now Turbo Agents have 2, Professional Agents have 5 and Executive Agents have 10 automated Skills. All tiers continue to enjoy unlimited manual Skills run via slash commands in Chat.

Original source

Day AI's native Gong integration pulls every transcript into the customer memory graph — linked automatically to the right contact, company, deal, and stage. Calls stop being recordings you have to go find. They become a queryable context that agents and reps can reason over alongside email, Slack, calendar, and meeting notes.

What it changes:

• For reps. Walk into every call already briefed. Every objection, commitment, and decision from the prior six conversations at your fingertips — no prep meeting, no scrubbing through Gong.
• For managers. Pipeline that reflects what's actually being said on calls, not what reps had time to log on Friday.
• For agents. Claude (or any agent) can draft a follow-up grounded in the actual conversation, with citations back to the moments that matter.

Gong captures the conversation. Day AI gives it memory.

Granola & Day AI

Day AI's Granola integration syncs every meeting note directly into the memory layer. Summaries, action items, decisions, and follow-ups flow automatically into the right deal, contact, and company. No manual tagging. No copy-paste. The meeting becomes durable institutional memory the moment Granola wraps.

What it changes:

• For reps. Your notes auto-update the deal record. The follow-up writes itself. The next teammate to touch the account inherits full context.
• For managers. Visibility into what actually happened in the meeting, without chasing anyone to log it.
• For agents. Every decision and next step becomes a structured fact agents can reason over — not loose text in a tool they can't see.

Granola captures what was said. Day AI remembers it — and acts on it.

Also shipped

Full meeting transcripts. Agents can now page through complete meeting transcripts, not just summaries. Ask a question about something said in minute 42 of a call and get an answer grounded in the actual exchange.

Page sharing via agent. Your agent can now set and update sharing permissions on pages directly — no bouncing out to settings. Just tell the agent who should see what.

Tabs + Affinity for MCP Client. Two new connectors are live. Day AI agents can now read and act across Tabs and Affinity alongside the rest of your stack.

Meeting recording import pipeline. Migrating from Chorus or Fathom? Historical meetings can now be imported directly into Day AI's memory layer — so the context you've already built doesn't get left behind.

Larger image and PDF uploads. Size limits raised on both. Drop in the full deck, the full contract, the full screenshot — Day AI handles it.

Email Attachments. You can now add attachments to emails sent from within Day AI, no more copy-pasting what your Agent drafted into Gmail.

Calendar visibility. All calendar events are now visible in Person and Organization record timelines, and your Agent can search across your team’s full historical calendar events instantly

Original source

Bring Linear, Notion, and HubSpot into any chat

Connect Linear, Notion, or HubSpot to Day AI and your agent can pull from them and take action in them, right inside the conversation you're already having.

Prep for a call by grabbing a Notion doc. Turn a bug from a meeting into a Linear ticket. Pull in your active pipeline from HubSpot and plan a next step for each deal.

Day AI now works with more of the tools you already use. And we're going to keep rolling out connectors.

Set it up in Workspace Settings 1 User Settings 1 Connectors.

Ask a question. See a chart. Drill down into the numbers

The analysis is more accurate because your agent isn't doing the math itself anymore. It writes Python, runs it against your actual data, and builds the chart from real numbers.

No estimation, no inference. And if you want to go deeper, click into the chart and see the records behind it.

You can also add charts to a page and share them with your team.

Search that keeps up, and full transcripts

Search counts are now accurate across tens of thousands of records.

Before, search capped out around 10,000 results, which meant counts were wrong for larger workspaces. Long meetings used to get cut off at the transcript level.

Now you can ask your agent for the full thing and get the whole conversation, paginated from start to finish.

Also shipped this week:

• Admins can now set up and configure agents for their teammates directly, from a new Agents tab in Workspace Settings
• Delete email drafts one by one or in bulk
• Every new agent now comes with a Morning Briefing skill out of the box, no setup needed

Original source

We built Day AI for early-stage founders. They need customer memory. The product works for them.

When we ran our PMF survey, one thing stood out: the larger the company, the more disappointed they would be if Day AI went away. Scale-ups, companies with real GTM teams, Gong deployed, Salesforce in place, the full stack, scored higher than we expected. We did not build for them. They adopted anyway, and they stayed.

That surprised us. It shouldn't have.

Why scale-ups need customer memory more

At five people, everyone knows what happened on every call. No one has to ask.

At fifty people, that breaks. At two hundred, it is completely gone.

More conversations than any individual can track. More teams who need to know not just what was said, but what to prioritize and what to do today. Sales, marketing, CS, and product all need an answer, every day.

That is where customer memory matters most. Not a place to store what happened. The thing your agents actually run on.

The problem with Gong data

Scale-ups already had the best call data in the world sitting in Gong. Almost none of it was available to the agents they were trying to run.

Gong does what it was built for: recording, transcription, coaching, forecasting. But the data was stranded from the broader workflow. So teams started building bridges by hand.

One person described his actual morning routine: pull the transcript from Gong, paste it into Claude, build the whole day from there. Others went further, assembling internal systems that pulled in Slack data, meeting transcripts, whatever they could access. Agentic workflows built by hand, one integration at a time.

They'd get most of the way there. Then they'd want email history. Under 100 employees, you can spin up a dev app and make it work. Over 100, you hit the Gmail API audit. Then you start thinking about permissioning: who can see which emails, which meetings, which conversations. That problem becomes more than a full-time job. The fun stops. The system gets shelved.

These are not startups hacking things together. These are companies that made serious investments in their GTM stack, chose to keep Day AI because customer memory became essential to it, and were doing manual work that should not exist.

The integration

One connection. Every Gong recording flows automatically into the customer memory layer.

Every morning, an agent reviews open opportunities, pulls the full call history, runs web research on what has changed at each account, and returns a prioritized list of who to reach out to, why, and a drafted message, before the first call of the day. Not a CRM view to scroll through. A plan to execute.

Marketing briefs campaigns using actual phrases customers used to describe the problem, not what someone on the sales team remembered to pass along. Product gets a weekly digest of every feature request that came up on calls, organized by frequency and segment. CS walks into renewals with a full brief: what was promised, what came up repeatedly, what is at risk.

The competitive battle cards that are always three months behind? An agent reads every call, pulls every competitor mention, and rewrites them automatically.

Gong owns the recording. Day AI is where it goes next, including Claude.

"I've been the Salesforce admin at every startup I've been at for fifteen years. Every single one. New company, same story.

The first time I tested Day AI's Gong integration I had one of those moments: 'oh, that's what this was supposed to feel like.' Sixty seconds to connect. Everything was just there, structured. The agents could actually use it."

— Chris Gadek, CEO, AdQuick

What this tells us

We did not want anyone finding this a year from now and wondering why no one told them. So we made it easy to try.

We were right that companies need customer memory. We just underestimated how much scale-ups need it, and that the customers who rely on it most would already have everything else in place.

Original source

We've been focused on making your assistant more capable and more proactive, so it can do more for you before you even ask. Here’s what we’ve shipped recently.

What’s new

Talk to your assistant directly from your inbox.

Reply to any skill email to ask follow-ups or adjust the output. Or start a fresh conversation by emailing [email protected]. No need to open Day AI to get things done.

Trigger custom skills after every meeting.

Build skills that fire automatically when a meeting recording summary is ready. Extract action items, update CRM records, send custom Slack summaries. This was one of our most requested features, and it's live for all user-created skills.

Re-enrich records with one click.

Click the wand icon on any person or org record and get research across multiple data sources, cross-referenced results, and filled in name, title, company, photo, LinkedIn, career summary, and more.

What’s improved

The Day AI SDK works like a REST API.

Write client.search(...) and client.createPerson(...) instead of thinking in MCP. Tool docs rebuilt from scratch to match the 49 tools actually in production.

Tables your assistant builds in chat are fully interactive.

Tables include live object chips with logos and one-click navigation to any record. Click straight through to what you need instead of searching separately.

Chat streaming is smoother and cleaner.

Fixed choppy streaming and message replays that some users reported. Tables animate row by row.

Sending email feels like it should.

After hitting Send, the editor transitions smoothly to show your sent message in its final form. The layout is cleaner, and transitions are smoother throughout.

Organization chips are clickable everywhere.

Org chips on person and opportunity details open the full org record in one click.

Invite teammates and provision their assistant in one step.

Billing admins see a new Assistant Tier dropdown right in the invite dialog. Pick a tier, send the invite, and the new teammate's assistant is ready when they arrive.

Meeting recording emails suggest what to do next.

Recording notifications include AI-generated conversation starters tailored to your meeting. One click takes you into a chat with your assistant, pre-loaded with context.

Your Default View saves your work.

The Default View on every table page is a real, persistent view. Rename it, save filters and sorts, share it with your workspace.

Right-click to copy any table cell.

Right-click any cell to instantly copy its contents to your clipboard.

Contact suggestions when composing email.

Start typing in To, CC, or BCC and see instant suggestions with name, photo, title, and company from your workspace contacts.

Meeting date visible in the recording sidebar.

No more cross-referencing the table view to figure out when a meeting happened.

Original source