Sachin Dev Duggal returns with SeKondBrain and Kemory, in a bid to make AI memory portable

Since the collapse of Builder.ai, Sachin Dev Duggal has been busy. Today he’s formally launching his new startup, SeKondBrain, whose first product, Kemory, aims to give users a single memory layer spanning AI systems, and hopes to be your first port of call for AI. He spoke exclusively to Pathfounders about his new startup.

By now we are all familiar with juggling multiple AI platforms. We might use ChatGPT for casual conversations. Claude for agents. Gemini for Google apps integrations. As a result, of course, our conversations, preferences and working contexts are being fragmented across these different systems.

Duggal thinks that this ‘accumulated context’ may ultimately matter more than the individual AI model being used to access it.

“If I think about how I use AI, and I use a bunch of them, every AI I use remembers a bit of me, but none of them remembers the same bit because it depends where you had what conversation,” Duggal told me over a call from his new home in Dubai. 

“We realise really quickly that whilst the AIs have bits of memory themselves, there's really not one unified memory layer across them. Secondly, you're kind of stuck, right? So if you're using ChatGPT for the last one and a half years, you've got so much history in it. How do you get your history out, and how do you move it around easily? Obviously, you can go to export, export the 10 gig file, but then how many times are you going to do that?”

“The minute I switch between one AI and the other, I’m basically the integration layer. I have to copy and paste things out and paste it somewhere else.”

“What if you want to just keep trying different AIs to use different things? And so that's why we created Kemory. It's the first product of SeKondBrain. It's the unified memory that you can plug any of the AIs into, and you can build agents on top of.”

That problem became the basis of his idea for Kemory. Duggal says that, in its first incarnation, SeKondBrain originally built shared memory for its own AI agents so they could work from a common pool of knowledge rather than repeatedly passing context between one another.

“We’re building agents like humans, and humans have a ‘bug’. Our minds can’t be connected,” he said. “We’re having these machines continuously talk to each other when really they shouldn’t have to talk to each other. They should be able to share memory.”

The resulting product is that Kemory combines three things: memories, chats and artefacts. Chats are visible AI conversations; artefacts are documents, images and other outputs; “memories” are contextual information derived from those interactions.

The service pulls these together and organises them into what SeKondBrain calls “namespaces”. During a live demonstration, Duggal showed me an account containing more than 18,000 memories spread across roughly 1,300 namespaces.

“What we’ve done is we bring memories, chats, and artefacts together,” he said. “The power here is I’m not having to now go figure out where did I have that conversation.”

Simply dumping an entire AI history into another model would, as many have found, be unwieldy and expensive, so SeKondBrain also uses what Duggal calls a “dialectic layer” to compress and synthesise previous conversations.

“You might have had 100 conversations, but all 100 conversations aren’t that relevant,” he said. “We’re compressing it down, synthesising it, so that when the next AI picks up, it doesn’t have to go through the 100 turns of conversation you had with ChatGPT. It gets a really well-synthesised summary.”

During the demo, Duggal moved a conversation from ChatGPT into Manus without manually copying the earlier context. Kemory retrieved the relevant history and passed it into the second system.

The product can also “enhance” a prompt with previous information, query a user’s accumulated memory directly, or continue a conversation elsewhere. It supports MCP, APIs and command-line interfaces, positioning it for developers and agents as well as consumers.

“We use it in coding for everything we’re building,” Duggal said. “The benefit is we can now have six coding agents running, and we don’t need them to communicate with each other.”

I asked whether OpenAI, Anthropic or Google could simply make Kemory unnecessary by improving their own memory features.

Duggal argues that individual model memory is not the real problem. Users, he believes, will continue to use several different AI systems because “no frontier model is going to be able to be good at everything”.

For Kemory to become redundant, he argues, major AI providers would effectively have to make their memory systems interoperable, something he regards as unlikely because it would weaken their walled gardens.

“I think sort of being ‘Switzerland’ in all of this is quite useful,” he said, speaking of Kemory. “What we’re simply saying is everyone’s giving us the memory. You can continue to work wherever you want.”

SeKondBrain is betting that OpenAI, Anthropic and Google may well make their own products ‘stickier’ by remembering more about a user, but Kemory can make that accumulated history portable and reduce the cost of switching between the models. 

And Duggal does not plan to stop there. He thinks Kemory could incorporate context from email, WhatsApp, Zoom transcripts, browsing history and cloud storage. But, for now, he says SeKondBrain is starting with AI.

How Kemory gets access to users’ existing AI histories is via a Chrome extension which Duggal says can pull previous conversations from services including ChatGPT, Claude, Gemini, Perplexity and Manus.

I asked what would happen if the platform tries to block access? Duggal said the extension works through the user’s own logged-in browser, making blocking it hard. Plus, he said Kemory stores chat history and memories encrypted.

“It’s yours. We never learn from it. We can’t read it,” he said, adding that memory is unlocked when queried or searched.

SeKondBrain also plans a locally-run ‘community edition’ for privacy-conscious users and organisations.

The initial target market is “prosumer and developer”. A free tier will include up to 2,000 memories and two AIs; the Pro version is planned at $20 a month, followed by a business edition.

Duggal said SeKondBrain currently has 13 staff across London, Athens, Berlin, Dubai and Delhi. He says it has so far generated revenue in the “mid hundreds of thousands” from design partners in legal, consulting and construction. He has funded the remainder of the build-out himself and says he expects to consider a structured financing round towards the end of the year.

But of course, the controversial story of Builder.ai’s collapse could well haunt his new vehicle.

Builder.ai was the AI-assisted software development startup he co-founded, which grew into a unicorn valued at more than $1bn before it entered insolvency in May 2025.

A Financial Times story alleged that an internal investigation at Builder.ai found evidence of potentially bogus sales and that previously reported revenues were subsequently revised down substantially. Before the collapse, lawyers for Duggal and Builder.ai said neither he nor the company had misreported sales figures.

The FT also reported that in Summer 2025 the FBI was investigating circumstances surrounding Builder.ai’s collapse and had approached a former senior employee about appearing before a grand jury. The current status of that reported investigation is unclear.

What is clear is that Duggal had left Builder.ai before the insolvency.

When I asked about the affair, he said: “There were mistakes made, built too fast, built too much. There’s a lot of water under the bridge.”

He also says he is allocating equity in SeKondBrain to former Builder.ai investors and employees.

“These folks invested in me, and they didn’t get the result they wanted, and I want to make sure there’s a path for them to get a tangible result at least out of this,” he said.

When I asked whether that amounted to a mea culpa, Duggal clearly wanted to put the issue in different terms, replying that there was a “moral responsibility”.

“People gave their time, people gave their efforts,” he said. “I want to make sure people get their money back.”

Make of that what you will.

But the legal tribulations Duggal faces are not yet over. He is also facing a separate legal matter in India concerning Nivio, the cloud-computing startup he founded before Builder.ai.

According to the Financial Times in April this year, India’s Directorate of Enforcement filed a prosecution complaint in 2026 alleging that Duggal was the “key beneficiary” of a scheme in which funds from the now-defunct Videocon Industries were siphoned and laundered through overseas entities.

The allegations concern transactions involving Nivio and related entities between 2008 and 2014, several years before Builder.ai was founded.

The FT reported that a court would need to accept the merits of the agency’s complaint for Duggal to be formally accused. 

For his part, Duggal has denied wrongdoing. His representative told the FT they could not comment on the substance before it was heard by a judge, but were confident “that these baseless charges shall be dismissed”.

These allegations by the Indian authorities are not findings of guilt, and the entire matter is totally separate from Builder.ai’s collapse. There is also no implication that Duggal or his associates have done anything illegal.

Meanwhile, for Duggal, SeKondBrain is clearly his potential ‘third act’, hoping that AI stays as multi-model as it’s increasingly becoming.

Duggal is also developing a “council mode” intended to let users ask several systems (potentially Claude, ChatGPT, Gemini, Perplexity and open-source models) the same question and receive a ‘synthesised’ answer.

He thinks the AI model, user interface and memory should no longer be mixed up.

“Today, the problem is the user interface, the memory, and the models are all tied together,” he said. “They all need to be, in my opinion, disintermediated.”

If he is right, users may eventually care less and less about which AI using at any particular moment, and much more about where their history, or even their ‘playlist’ is. Hence, perhaps why SeKondBrain leans into the idea of the ‘second brain’ so many people have tried to build out themselves. 

“It’s a passport versus a visa where all the AIs take your luggage,” Duggal told me.

It’s not entirely clear if Duggal will have enough of a ‘moat’ around what Kemory is. If he can build a Chrome extension to mix AI models, surely someone else could? 

Perhaps, as with so many of his previous projects, Duggal’s prodigious talents as a highly capable, technically proficient, entrepreneur, storyteller and, let’s be honest, salesman may be the main ‘moat’ others will find hard to breach.