Current AI launches Alpha Chat to assemble a public open-source AI stack
CEO Ayah Bdeir is applying her littleBits playbook to a nonprofit coalition backed by more than EUR400 million in commitments.
By Ryan Merket ยท Published
Why it matters
Bdeir is testing whether public and philanthropic capital can support a usable AI stack with open components and community-controlled data. Success would give builders an alternative to relying on one proprietary provider from model to interface.
On July 9th, Current AI CEO Ayah Bdeir launched Alpha Chat, a no-login chatbot that combines components supplied by open-source organizations into a single product. Current AI plans to spend the next 12 months extending that prototype into an integrated public AI system spanning models, safety tools, compute and applications.
The release, reported by TechCrunch on July 19th, gives concrete form to an organization that began in February 2025 as a funding and coordination vehicle. Martin Tisne founded Current AI during the Paris AI Action Summit after years working at the boundary between philanthropy, government and technology. Tisne helped establish the Open Government Partnership with the Obama White House and later led digital-governance investments at Luminate.
Bdeir joined Current AI on January 20th after leading Mozilla's AI strategy. Her appointment put a product builder in charge of Tisne's institutional coalition, and Current AI's releases since June show the resulting shift: direct grants, a map of open-source technical components, a handheld AI device and a working chatbot.
Bdeir returns to an open-hardware playbook
Bdeir studied engineering and sociology at the American University of Beirut before joining the Computing Culture group at the MIT Media Lab. She later became a Creative Commons fellow, helped develop the open-hardware movement and co-authored the first Open Source Hardware Definition.
In 2011, she founded littleBits, which sold modular electronic building blocks designed to let children build devices without conventional engineering training. Bdeir's biography says littleBits generated more than $150 million in lifetime revenue and reached 20,000 schools before Sphero acquired it in 2019. The acquisition price was never disclosed.
That record explains Current AI's approach under Bdeir. She is treating openness as a product and distribution problem. Publishing model weights or code is insufficient if users still need proprietary interfaces, closed computing services or specialist knowledge to put the pieces together.
In her January announcement, Bdeir defined access as technology that communities can understand, afford and modify. Current AI's job, in her telling, is to connect existing components and fund the pieces that commercial developers have little incentive to build.
Alpha Chat tests coordination before capability
AI Potluck is Bdeir's attempt to organize that work around a shared product roadmap. Contributors include Hugging Face, Mozilla.ai, OpenMined, Public AI, the Algorithmic Justice League, MIT Media Lab's Advancing Humans with AI group and the LUMI supercomputer.
Current AI says Alpha Chat was assembled in about eight weeks. The prototype uses Apertus as its base model and runs on public compute. Current AI also says the training pipeline is open and the underlying data was collected with consent. Alpha Chat avoids login requirements and includes a feature that shows users which components sit behind a response.
The product is early. Current AI has not published usage, retention, response-quality benchmarks or operating costs. Alpha Chat therefore establishes that independent contributors can integrate their work into a functioning interface. It does not establish that the resulting service can match proprietary systems on reliability, speed or sustained demand.
Current AI has set June 2027 as the deadline for a working prototype of the broader stack. The organization says success will eventually include adoption by startups, public agencies and civil-society groups, although it has not set public numerical targets.
A map for deciding what to fund
Before launching Alpha Chat, Current AI published its Open Source AI Gap Map on July 1st. The project assessed 24,626 open-source artifacts and examined 421 in depth: 266 software tools and libraries, 85 models, 50 datasets and 20 hardware projects produced by 228 organizations.
Current AI scored the selected projects for openness, capability and adoption. The remaining roughly 24,400 entries have not received detailed scores, a limitation that makes the first version more useful as a directory and funding guide than a comprehensive measurement of open-source AI.
The map serves Current AI's financial strategy. It is meant to identify layers where capable projects lack distribution, where widely used tools remain only partly open, and where too few maintained alternatives create technical dependencies. Bdeir can then direct grants or contributors toward specific gaps instead of funding another general-purpose model.
The first grants reveal Current AI's priorities
On June 19th, Current AI announced $3.2 million for four organizations working on language data, cultural archives, local AI deployment and auditing.
Masakhane received $2.25 million to expand datasets covering more than 50 African languages, including material for health, agriculture and education. The Institute for Worldmaking in Lebanon and Brazil's Portal sem Porteiras each received $285,000. The African Internet Rights Alliance received $290,000 for AI audit tools.
Those itemized awards total $3.11 million, which is $90,000 below the announced $3.2 million cohort. Current AI's announcement does not explain the difference. The allocation also concentrates roughly 72% of the listed awards in Masakhane, making African-language data the clearest priority in the first cohort.
Current AI says Portal sem Porteiras will keep models and data within Indigenous territories, while the Institute for Worldmaking will create machine-readable Arab cultural databases governed by participating communities. These projects remain works in progress, and Current AI has not published delivery dates or outcome measures for the grants.
Suno Sutra moves the thesis onto a device
Bdeir's hardware background is most visible in Suno Sutra, a pocket-sized prototype developed with Bhashini, the Indian government's language technology program. Current AI and Bhashini unveiled it on February 20th at India's AI Impact Summit.
The device combines a camera, microphone, speaker and screen. It runs vision, speech-recognition, translation and text-to-speech models locally, allowing it to describe a user's surroundings in India's 22 official languages without an internet connection. Current AI lists 8GB of memory, 128GB of storage and estimated battery life of one to three hours.
Current AI has released software, hardware designs and build instructions, and opened the VYOMA Innovation Challenge in June with more than $100,000 in support for developers. Production volume, unit cost, distribution plans and active use have not been disclosed. Suno Sutra should be read as a developer platform rather than a shipping consumer device.
Commitments are not operating cash
Current AI describes itself as an independent French nonprofit association backed by more than EUR400 million in commitments. Its founding partners included the French government, AI Collaborative, Google, Salesforce, the Ford Foundation, the MacArthur Foundation and the Patrick J. McGovern Foundation.
Bdeir told TechCrunch that these organizations are funders rather than investors. Current AI has no equity valuation, and its backers receive no ownership stake comparable to a venture round. The organization aims to mobilize $2.5 billion over five years.
Current AI has not disclosed how much of the committed capital has been received, its annual operating budget, current headcount or the amount available for future technical work. Those omissions matter because maintaining a full AI stack requires recurring spending on compute, security, engineering and support after the initial prototypes are released.
Bdeir's central bet is that coordinated public funding can support useful AI infrastructure without one owner controlling the data, product and computing layer. Alpha Chat gives that bet an interface. Suno Sutra gives it hardware. The Gap Map and grant program supply a mechanism for deciding what gets built next. Current AI's next test is keeping those parts maintained, integrated and useful long after the launch cycle ends.