Sesame AI is turning its voice companion into an assistant that does work while you are not talking to it. A test build of the iOS app, shared on 30 September 2026 by the leak-tracking account TestingCatalog, shows a hidden Activity tab with sections for Tasks, Schedules and Skills, plus connectors for Gmail, Google Calendar and Google Drive. The same post says the thinking effort of the underlying model “will be adjustable”, and that paid plans are likely.
None of this has been announced by the company. What exists is a set of screenshots, a short description and two earlier leaks that show where the product has been heading since its public launch in May. Read together, they describe a clear shift: from a voice you chat with to an agent you hand work to.
This article sets out what the screenshots show, how the connectors and skills work, how the product compares with larger assistants, and what a business should check before connecting a work account. For background on the voice technology, see our overview of Sesame’s Conversational Speech Model and our wider AI models and tools hub.
Table of contents
- What Sesame AI Is Building Now
- Who Sesame Is
- The Sesame AI Timeline
- How Sesame AI Connectors Work
- Sesame AI Skills Follow a Familiar Pattern
- Tasks and Schedules in Sesame AI
- How Sesame AI Compares With Bigger Assistants
- What Paid Plans Could Look Like
- Privacy Questions for Sesame AI Connectors
- What Sesame AI Means for Businesses
- What We Still Don’t Know About Sesame AI
- Sesame AI FAQ
- References and Further Reading
What Sesame AI Is Building Now
TestingCatalog’s post opens with a single line: “Sesame is building support for Connectors, Skills, Scheduled tasks, and task management features!” Two screenshots from the app back it up. The header shows the agent Miles, with three tabs beneath it: Activity, Apps and Agents.
A hidden Activity tab
The Activity tab opens with the line “Miles’s here to help. Hand off a task and Miles gets to work in the background.” Below it are three sections. Tasks is empty, with the prompt “Ask Miles to plan a trip or research something. You’ll see them here.” Schedules reads “Nothing scheduled” and suggests asking for “a daily briefing tomorrow or every morning at 7am”. Skills shows one test entry, described simply as “When user says test”.
An Apps tab with three connectors
The Apps tab lists “Installed apps”, each labelled “by Sesame”. Gmail promises to help you “Stay on top of your inbox.” Google Calendar will “Manage your calendar effortlessly.” Google Drive can “Create and read files from your drive.” All three are Google services, which matters for any business that runs on Microsoft 365, as we cover below.
Thinking effort and paid plans
Two further details come only from TestingCatalog’s text. First, “Thinking effort for the underlying model will be adjustable as well.” Second, “Currently, the app is free; however, Sesame will likely introduce several paid plans soon. Pricing and feature set are yet unclear.” The author adds that they still use Sesame AI as their “default voice assistant”.
| Tab | What the 30 September build shows | Source |
|---|---|---|
| Activity: Tasks | Background jobs such as trip planning or research | Screenshot |
| Activity: Schedules | Recurring jobs, such as a 7am daily briefing, with a Create button | Screenshot |
| Activity: Skills | Custom instructions triggered by a phrase, with a Create button | Screenshot |
| Apps | Gmail, Google Calendar and Google Drive, all “by Sesame” | Screenshot |
| Agents | Tab visible; contents not shown | Screenshot |
| Settings | Adjustable thinking effort; paid plans expected | Post text only |
Who Sesame Is
Sesame is a start-up co-founded by Oculus co-founder Brendan Iribe and others from the virtual reality company Meta bought. Its pitch has always been voice that feels human. The hard part of that pitch is not only natural language processing, working out what someone means, but timing: pauses, interruptions and tone, the things that make a phone call feel like a conversation.
A demo that unsettled people
In late February 2025 Sesame released a research preview with two voices, Maya and Miles. It spread quickly. Ars Technica’s headline said the “eerily realistic” demo “sparks amazement and discomfort online”, and another outlet reported the voices were “freaking people out”. According to Sesame investor Sequoia, the voices were used by more than a million people in the first few weeks.
Open model, big funding round
In March 2025 the company released its base speech model, CSM-1B, publicly. TechCrunch later reported that Sesame had raised a $250 million Series B from Sequoia and others. That is a large budget for a company whose only public product is a free app, which is one reason paid plans were always coming.
The iOS app arrives
On 28 May 2026 Sesame AI released a public preview of its iOS app in 39 countries. It offers four agents, Maya, Miles, Simone and Charlie, “each of which have their own distinct voice, personality, point of view, and memory”, TechCrunch reported. Beta features included search cards with images, notes, a texting mode, deep dives and an incognito mode that saves nothing to memory. An Android preview was promised for later.
The glasses behind the app
The app is not the end goal. TechCrunch said Sesame expects to launch “intelligent eyewear” in 2027, and that the company hinted its agents would “later be able to take action on your behalf”. The features in the new build are the first concrete sign of that second promise.
The Sesame AI Timeline
The table below puts the public milestones in order. The day counts are our arithmetic.
| Date | Milestone | Gap |
|---|---|---|
| Late Feb 2025 | Research preview with Maya and Miles | – |
| Mar 2025 | CSM-1B base speech model released | Weeks |
| 28 May 2026 | iOS public preview, four agents, 39 countries | About 15 months |
| 14 Jul 2026 | Calendar and Email connectors and a custom skills menu spotted | 47 days |
| 30 Sep 2026 | Activity tab, three Google connectors, thinking effort, paid plans expected | 78 days |
| 2027 | Intelligent eyewear expected | – |
What the pace suggests
It took 125 days to go from a talking app to a build with background tasks. The July leak showed connectors as a settings page; by September they had become “Installed apps” beside a task list. That is the normal path from a feature flag to a product surface, and it suggests a launch is closer than the July screens implied. It does not give a date.
How Sesame AI Connectors Work
Connectors are the permissions that let an agent read or change data in another service. They are what turn a conversation into an action.
July: Calendar and Email, per agent
TestingCatalog’s first leak, on 14 July, showed an “Integrations” screen with a separate tab for Maya, Miles and Charlie. Each agent had its own Calendar and Email switches. The descriptions were broad. Calendar could “Create, search, view, open, update, and delete events in Google Calendar.” Email could “Search, read, send, and delete Gmail messages.” The post said users “will be able to provide Sesame agent an access to read and potentially operate these apps”.
September: three apps, one new
By 30 September the list had become three installed apps, with Google Drive added, and plainer marketing descriptions. The change in wording does not tell us whether the permissions changed. It is safest to assume the July scopes still apply until Sesame AI publishes its own documentation.
Number of actions named in each connector’s description (counted from the leaked screens)
“By Sesame” hints at a directory
Every connector in the September build is labelled “by Sesame”. That label only makes sense if connectors by someone else are planned. An app directory with third-party connectors is the pattern ChatGPT and Claude both followed, and it would be the natural next step.
Why voice plus tools is different
In July TestingCatalog explained why it kept using Sesame AI: “it is both duplex but also has access to tools (reminders and note taking)”. Duplex means you can talk over it and it keeps up, like a phone call. The same post noted that OpenAI’s GPT-Live-1 voice model “lacks tool access at this moment”. A natural voice that can also act is the gap Sesame AI is aiming at.
Sesame AI Skills Follow a Familiar Pattern
The July leak included a “New Skill” form. It is worth reading closely, because it copies a format that has become standard across AI assistants.
Three fields
The form has a Name field, with the example “plan-birthday-party”. It has a Metadata field for “When and why to use this skill”, with the note “Short summary of what the skill is for — always visible to the agent.” And it has an Instructions field, “Loaded on demand when the agent invokes the skill.”
The Agent Skills standard
That is the design of the open Agent Skills specification, which grew out of Anthropic’s work on Claude. There, a skill has a name and a description that are “loaded at startup for all skills”, and a body of instructions “loaded when the skill is activated”. Names may contain only lowercase letters, numbers and hyphens, which is exactly the shape of “plan-birthday-party”.
| Element | Sesame’s New Skill form | Agent Skills specification |
|---|---|---|
| Identifier | Name, e.g. plan-birthday-party | name: lowercase, numbers, hyphens, up to 64 characters |
| When to use it | Metadata, “always visible to the agent” | description, loaded at startup (about 100 tokens) |
| What to do | Instructions, “loaded on demand” | Body, loaded when activated (under 5,000 tokens recommended) |
| Extra files | Not shown | Optional scripts, references and assets |
Why the pattern matters
Loading only short descriptions until a skill is needed keeps the agent fast and cheap, which matters more for a real-time voice than for a text chat. It also means skills written for one assistant may be easy to move to another. For users, a skill is a saved routine: say a phrase, and the agent follows your written steps.
A skill in practice
Here is how a user might fill in the form. Name: “weekly-wrap”. Metadata: “Use when I say ‘wrap up my week’ on a Friday.” Instructions: check Google Calendar for next week’s meetings, search Gmail for unanswered messages from the past five days, and read out the three most urgent items, then offer to draft replies. Nothing in that routine is new. What Sesame AI adds is the ability to trigger it by voice, mid-conversation, and hear the answer rather than read it.
Tasks and Schedules in Sesame AI
The Activity tab is the biggest change in the build, because it separates the conversation from the work.
Hand off, then walk away
“Hand off a task and Miles gets to work in the background” describes an agent that keeps working after the call ends. The examples, planning a trip and researching something, are multi-step jobs that take minutes, not seconds. Results would appear in the Tasks list rather than being read out immediately.
Briefings on a timetable
Schedules add time. “A daily briefing tomorrow or every morning at 7am” implies a job that runs at a set hour and, presumably, calls on connected apps: today’s meetings from Calendar, overnight mail from Gmail. ChatGPT added scheduled tasks in January 2025, so the idea is not new. Doing it with a voice that briefs you conversationally would be.
Thinking while talking
Sesame AI has already solved a related problem. At launch, the company described “an inherent tension between replying quickly and taking the time to compose thoughtful responses”, and said its agents run searches in parallel while speaking. Adjustable thinking effort extends that trade-off to the user: quicker answers, or slower and more careful ones.
How Sesame AI Compares With Bigger Assistants
The individual features are familiar. The combination, on a voice-first product, is less common.
What others already offer
ChatGPT has app connectors, scheduled tasks and a plugin directory. Anthropic’s Claude popularised the skills format. Meta has just added app connectors to its Muse assistant, as we reported in Muse for Small Business. Sesame is not first to any of these.
Where Sesame AI differs
Its advantage is the voice. The big assistants are text products with a voice mode added; Sesame AI is a voice product adding tools. For people who want to manage a calendar while driving or cooking, that difference in starting point could matter more than any feature list.
Where it is behind
It supports three connectors from one vendor, launched on iOS first, and has no published pricing. Nothing in the leaks points to a business tier or admin controls. Larger assistants publish admin controls, data retention terms and enterprise agreements. Until Sesame does the same, it is a consumer product.
What Paid Plans Could Look Like
TestingCatalog is clear that “pricing and feature set are yet unclear”, so anything more specific is guesswork. Two things are reasonable to expect.
Compute is the likely lever
Background tasks, scheduled briefings and deeper thinking all use more computing power than a chat. Charging for heavier use, with a free tier for conversation, is how most AI companies price these features. Adjustable thinking effort fits neatly into that kind of ladder.
Do not plan around free
The TechCrunch report on the May launch said the full experience was free “for the time being”. Anyone building a routine around Sesame AI’s background features should expect some of them to move behind a paywall.
Privacy Questions for Sesame AI Connectors
Giving any assistant access to email and calendar is a serious decision. Voice adds a few extra wrinkles.
Send and delete are powerful rights
The July descriptions included “send” and “delete” for Gmail and “delete” for Calendar events. An assistant that mishears a spoken instruction could act on the wrong message. Look for confirmation steps before destructive actions.
Background jobs need a record
A task that runs at 7am while you sleep should leave a log of what it read and what it changed. The empty Tasks list suggests results will be shown; whether every action is recorded is not visible yet.
Separate agents, separate access
The July screen gave each agent its own connections. That is good practice, but it also means checking permissions agent by agent rather than once.
Questions to ask before connecting a work account
Where are transcripts and connector data stored, and for how long? Is the data used to train models? Can you revoke access instantly? Does the incognito mode also cover connected apps? Sesame AI has not answered these publicly for the new features.
What Sesame AI Means for Businesses
For most organisations this is a product to watch rather than deploy.
A glimpse of voice-first work
Hands-free briefings and spoken task hand-offs suit mobile staff, site workers and busy managers. If Sesame AI delivers a natural voice with reliable actions, it will push larger vendors to improve theirs.
A Google-only start
All three connectors are Google services. Organisations on Microsoft 365 have nothing to connect yet, and should not route work mail through a personal Google account to try it.
Wait for the terms
Before any business use, wait for published pricing, a data processing agreement and admin controls. Until then, treat it as a personal productivity experiment on personal accounts.
What We Still Don't Know About Sesame AI
Leaks show what is being built, not what will ship. Four questions remain open.
When it launches
Neither leak gives a date, and Sesame has not commented. Hidden tabs in test builds sometimes ship within weeks and sometimes never. The step from a settings page in July to a named tab in September is encouraging, but it is not a commitment.
Which agents get the tools
The July screen showed connector switches for Maya, Miles and Charlie, but not Simone. The September screenshots show only Miles. Whether every agent will be able to run tasks, or only some, is unclear.
What the price covers
TestingCatalog expects “several paid plans”. It does not say whether connectors, skills and schedules will be free with limits, paid only, or split between tiers. The answer will decide whether Sesame AI’s new features reach most of its users or only a paying few.
How the data is handled
The leaks say nothing about where connector data is processed, how long task results are kept or whether any of it is used to train models. For a product that listens as well as reads, those answers matter as much as the features.
Sesame AI FAQ
What is Sesame AI?
Sesame is a start-up co-founded by Oculus co-founder Brendan Iribe that makes realistic conversational voice agents. Its iOS app offers four agents: Maya, Miles, Simone and Charlie.
What new features is Sesame AI building?
A test build shows an Activity tab with Tasks, Schedules and Skills, connectors for Gmail, Google Calendar and Google Drive, and adjustable thinking effort. None of it is officially released.
Is Sesame AI free?
The app is currently free. TestingCatalog expects “several paid plans soon”, with pricing not yet known.
What are Sesame AI skills?
Saved instructions with a name, a short description the agent always sees, and detailed steps loaded only when needed. The design matches the open Agent Skills specification.
Which devices does Sesame AI run on?
The public preview launched on iOS in May 2026, in 39 countries. Sesame said at the time that an Android preview would follow.
References and Further Reading
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
