Katalyst is the AI sales agent for teams on Salesforce. Hang up a call and it's already done: notes summarized, records created, fields updated, follow-up drafted, next step set. It runs 24/7, reading every call, email, and calendar, surfacing the right signals, prepping you for each meeting, and prompting you to act: follow up here, this one’s slipping. What's new: AI Resolution on every account, meeting recorder, hygiene scores, deal patterns. Built for enterprise sales teams.
Free Options
Launch Team / Built With
AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
@realanupreet This comes up with enterprise teams and it's exactly the kind of infra requirement we take seriously. Rather than answer it loosely here, let me get you specifics, what's your setup?
@realanupreet To add, routing critical data through your own AI infrastructure is a legitimate enterprise requirement, not an edge case, so it's worth getting the specifics right rather than a quick yes or no. Happy to dig into your setup alongside Div, exactly the kind of thing we'd rather scope properly than hand wave.
How does Katalyst prevent inaccurate AI-generated insights from affecting pipeline forecasts or CRM data?
@robert_dimla Every update shows its reasoning and needs rep approval early on, and high-stakes fields like forecast category can stay approval-only, so a bad read never silently hits your pipeline.
@robert_dimla To add, this is the guardrail we cared about most, nothing writes silently. Every update carries its reasoning and passes through rep approval, so an inaccurate read gets caught before it ever touches your forecast rather than after. Accuracy you can see beats accuracy you're asked to assume.
This is a smart approach to the "AI does the boring CRM work" problem.
Since Katalyst is auto-creating records and updating fields based on
call/email reasoning, how are you handling cases where the AI
misreads a signal — like marking a deal as "slipping" when it
actually wasn't, or auto-filling a field with the wrong next step?
Curious if there's a review/undo layer before changes go live in
Salesforce.
@amjad_shaik Yes, changes surface for rep approval before they land, each with its reasoning, so a misread like a wrongly flagged "slipping" deal gets caught upfront, not after. And anything it writes automatically is logged in the activity history, so it's reviewable and reversible.
@divyansh_lohia That's a solid design — surfacing low-confidence writes as suggestions
rather than auto-committing them is the right call, and having
everything logged/reversible closes the loop nicely. Makes sense why
you built it that way coming from watching reps lose hours to this
exact busywork at Datadog. Following along, curious to see how this
holds up once you're past enterprise pilot and into messier mid-market
data.
@amjad_shaik One thing to add. The agent also knows when it isn't sure. If confidence is low, it never auto-writes anything, it just parks it as a suggestion for the rep to look at. And if a wrong value ever does get accepted, fixing it is one inline edit in Katalyst, which syncs straight back to Salesforce. So a misread costs the rep a few seconds, not a data cleanup project.
What's your retrieval architecture? Are you using vector search, structured CRM data, or both?
@shreyans_jain20 Both. Structured retrieval on the CRM objects, vector search over calls, emails, and calendar. The part we obsess over is grounding: every write traces back to the exact source it came from, so a field update can be audited to the specific call or email. Retrieval that can't cite itself is a non-starter when you're writing to someone's CRM
@shreyans_jain20 Short answer, both. The Salesforce data we treat as structured truth and query directly, no embedding, because you want exact field values. Calls, emails, and calendar run through vector search so we pull the right context per deal. One gives you the state, the other gives you the story
The "Do This Now" section is probably my favorite part. How does Katalyst decide what deserves my attention first?
@saksham_diwan Glad that's the one that stuck. It ranks by where your time actually moves the needle: a mix of [deal value], [risk of slipping], and [fresh signals from calls and emails]. Less "here's everything," more "do this next
@saksham_diwan Glad that's the one that stuck. It ranks by where your time actually moves the needle: a mix of [deal value], [risk of slipping], and [fresh signals from calls and emails]. Less "here's everything," more "do this next
Deals evolve over months. Does the AI understand that a champion from March might not matter anymore in July, or is every interaction treated independently?
@diksha_sonthalia Love this question, it's exactly the problem we obsessed over. No, interactions aren't treated independently, that would fall apart on long deals like yours. Every call and email gets added to the deal's memory, so each new conversation is understood in light of everything that came before it, plus where the deal stands today. And your champion example touches a principle we baked right into the pipeline: facts age well, urgency doesn't. What someone told you in March is still remembered as context, but stale urgency gets dialed down instead of being pushed like nothing's changed. So the AI carries the whole story with it, but what it nudges you on reflects where the deal actually is in July. Would genuinely love to hear how it does on one of your longer deals!
@diksha_sonthalia Great question, and no, every interaction isn't treated independently, that would break on long deals. Each call and email joins the deal's memory, so July is understood in light of March. The principle we baked in: facts age well, urgency doesn't. Your March champion stays as context, but stale urgency gets dialed down rather than pushed like nothing changed.
As someone working in a Communications and Strategy space, does Katalyst account for constant changes and updates to a singular project or task? How does it take into consideration for the different asks from stakeholders at various points in time?
@anagha_nair Love this question. Short answer, yes: Katalyst treats a deal as a living thing, not a one-time entry, updating off every new call, email, and meeting, and preserving each stakeholder's ask with attribution instead of overwriting. Curious what you're picturing though, since we're Salesforce-native today, are you thinking sales opportunities or broader project and stakeholder tracking?
@anagha_nair Really thoughtful question. Katalyst is built specifically around sales deals rather than general projects, but the problem you're describing, one long-running thing that keeps changing shape as different people weigh in, is exactly what a deal is, so it's the core thing we handle. The way it works: every new conversation gets read against everything that came before it, not in isolation. So when a stakeholder changes an ask in week six, the AI sees that against what was said in week one, updates the record to reflect the new reality, and queues the follow-ups that fall out of the change. Nothing is ever a frozen snapshot, the current state simply evolves as the conversations do.
And each ask stays tied to the person who made it, so when three stakeholders want three different things at three different times, that history lives on the deal rather than in someone's memory. One principle we baked in that you'd probably appreciate from a comms angle: facts age well, urgency doesn't. Old information stays as context, but the AI won't keep pushing last month's priority as if nothing has moved since. If your work lives in Salesforce, this would translate directly. If not, the honest answer is we're deal-shaped today.