Snap launched Specs Intelligence on Wednesday, an AI assistant that connects to Gmail, Slack, and other accounts to anticipate tasks and surface information before the user asks. The service is available in preview on iOS starting September 16, 2026, with a waitlist opening for a fuller early-access experience on Mac. Snap is positioning the product as an "anticipatory AI service" rather than a chatbot, though users can also converse with it directly.
Specs Intelligence ships alongside Specs, Snap's first consumer pair of augmented reality glasses. On the glasses, the assistant can pull personal content and place it "directly into view" when relevant — for example, delivering a daily briefing before work or surfacing flight, hotel, and dinner details during a trip. The mobile preview is the same underlying service without the AR surface, though Snap declined to specify on the record how the iOS preview differs from the full Mac release.
The pitch is behavioral: the assistant learns goals, routines, and relationships from connected apps, then acts without a prompt. Before a meeting, Snap says it can highlight decisions to make and questions to ask. Ahead of a trip, it can bundle logistics and flag a work deadline that lands during the visit.
“SPECS Intelligence builds an understanding of your goals, priorities, relationships, and routines from the apps and tools you choose to connect”— Snap, company statement
Key facts
- 01Snap launched Specs Intelligence in iOS preview on September 16, 2026, with a Mac early-access waitlist opening simultaneously.
- 02The assistant runs on a proprietary mix of US-hosted open-source models and local LLMs, according to Snap spokesperson Cassie Bumgarner.
- 03Specs Intelligence connects to Gmail, Slack, and other accounts to surface briefings, travel plans, and meeting prep without a prompt.
- 04Snap says personal content from connected accounts will not be used to train its models or serve personalized ads.
- 05The assistant ships alongside Specs, Snap's first consumer augmented-reality glasses, which can place personal content directly into view.
The competitive frame is direct. Meta rolled out Muse earlier this year as part of its Meta One bundle starting at $7.99, and Google's Gemini Spark occupies similar territory. All three products chase the same premise — an assistant that reads across a user's inbox, calendar, and messaging to act on context — and all three depend on users granting deep access to personal data.
Snap says the assistant runs on "a proprietary combination of open-source models hosted in the US alongside local LLMs." The company has not disclosed which open-source models it uses or how the local-model split works in practice. That mix is a departure from rivals: Meta's Muse relies on Meta's own Llama family, and Gemini Spark runs on Google's stack.
Snap spokesperson Cassie Bumgarner said the composition will change over time. That flexibility gives Snap room to swap in stronger open-source models as they ship, but it also means the assistant's behavior — and its failure modes — will drift without a fixed version to benchmark against.
“We regularly update the mixture of models as we continue iterating towards the best version of SPECS Intelligence”— Cassie Bumgarner, Snap spokesperson
Privacy is the harder sell. Snap says it cannot view or access personal content processed through Specs Intelligence, and that data from connected accounts will not be used to train or fine-tune its models or to target ads. That is a stricter commitment than several competitors have made, and it is a necessary one: Specs Intelligence is asking users to hand over Gmail and Slack access to a company best known for a disappearing-message app aimed at teenagers.
The trust gap is real. Assistants with broad account access have unnerved users in past rollouts, and the value of anticipatory suggestions depends on the model reading the right signals from a large surface area of personal data. A single wrong suggestion at the wrong moment — surfacing a private email during a shared AR session, or misreading a meeting's stakes — is the kind of failure that anchors user perception for a long time.
Snap also has to prove the assistant works as advertised outside a demo. Anticipatory AI is easy to script and hard to deliver: the model has to know when to speak up and, more importantly, when to stay quiet. Rivals have struggled with both halves. Whether Snap's open-source-plus-local-LLM stack produces better judgment than Meta's or Google's vertically integrated models is the question the preview will answer.
Specs Intelligence is Snap's most serious bid to matter in AI beyond Snapchat's My AI chatbot, and it is tied directly to hardware the company needs to sell. If the assistant lands, it gives Specs a reason to exist beyond novelty — a wearable that quietly runs a personal operating system across the user's day. If it doesn't, Snap has spent scarce engineering budget building a fourth entrant in a category where Meta and Google already have distribution and a two-year head start.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
