We are a proud sponsor of UK AI Agent Hack EP.5
We are backing the builders pushing open source AI into production. Our track challenges you to solve enterprise memory governance at scale — and every prize tier puts real product, not just cash, in your hands.
Open source AI only wins if the people building it can ship. So we are putting real product, real credits, and real jobs in the hands of the builders who show up.
Real product in the hands of builders, not just cash.
Every tier includes credits for BasedAPIs, so you keep building on the same infrastructure you used to compete, plus access to Hirebase so you can actually run your project without hiring a team around it.
And because open source AI needs more good people building it, every participant is eligible for a fast-track interview for open roles at BasedAI. We are actively hiring, and a hackathon is a better signal than any resume.
Enterprise memory governance at scale
Enterprise agents produce higher quality, more context-aware work when they share an organizational memory. The hard part is giving many agents access to a persistent shared memory without leaking information across boundaries — organizations, teams, workspaces, projects — at scale.
Track requirementBuild with open-weight models. Any model in your stack must be open weight — the kind you can serve yourself on BasedAPIs — with no closed, proprietary model in the loop.
The challenge
Build a permission-aware memory layer that does all of the following.
- 1Stay synchronized with source ACLs under concurrent updates.
- 2Enforce access at the retrieval layer, not the application layer — and without an LLM call in the permission path. Classification may use an LLM at write time; final enforcement must be deterministic.
- 3Produce audit logs that meet regulatory standards.
- 4Hold sub-200ms latency for permission checks at P99.
- 5Govern derived memory by lineage: a summary, embedding, or note inherits its sources' constraints, and revoking a source propagates to the derivative.
Bonus challenges
- Temporal access rules — leadership call notes that unlock after 30 days.
- Query-time inference prevention — detect when retrieved context leaks information across permission boundaries.
What we judge
- Security correctnessFalse-negative rate under 0.1%, derived-memory permissions that match their source ancestry, and resilience under adversarial red-teaming.
- Latency at scalePermission checks that stay O(1) or O(log n) in roles, under 50ms at P50 and 200ms at P99, across growing tenants and documents.
- AuditabilityEvery retrieval decision logged, so you can reconstruct who saw what, and when, for any document.
The infrastructure your prize credits run on.
The same inference infrastructure your prize credits run on. Build your agent's memory and reasoning on it during the hackathon, then keep going on the credits you win.
Operational capacity on demand, so a winning prototype can become something you actually run — not a repo that stalls the week after demo day.
BasedAI engineers on the floor with you.
Our team is mentoring the track in person across the build. Bring them your hardest architecture questions.
Hands-on sessions with the BasedAI team.
Two workshops during the build week. Drop in to go deep on the track with our engineers.
- Monday 29 June3:00 – 4:00 PM
- Tuesday 30 June4:00 – 5:00 PM
All times BST · Imperial College London
From registration to demo day.
- 01
Register for the event
Sign up on the official UK AI Agent Hack EP.5 page. Registration and logistics are run by the organizers.
- 02
Pick the BasedAI track
Choose Enterprise Memory Governance at Scale and read the full brief for requirements, the evaluation framework, and prior art.
- 03
Build the week of the event
Build at Imperial College London, 28 June – 4 July. Use BasedAPIs for inference and lean on mentors during the build phase.
- 04
Submit at demo day
Ship your repo, a demo, and a short writeup of the problem, the user, and the outcome. Judging happens at demo day on 4 July.
Before you register.
The essentials on eligibility, prizes, and how the BasedAI track fits into the event.
Who can enter the BasedAI track?
Anyone registered for UK AI Agent Hack EP.5. You can compete solo or as a team — the track is open to every participant at the event.
Do I have to build on BasedAI products?
You do have to build with open-weight models — any model in your stack must be open weight, with no closed, proprietary model in the loop. Beyond that the track rewards real, working systems over any specific stack, but the prizes are BasedAPIs credits and Hirebase access.
What do the prizes actually include?
Every tier is BasedAPIs credits plus Hirebase access. First place is $2,000 in credits and a year of mid-tier Hirebase; second and third scale down from there; three honorable mentions get $100 in credits each. Full breakdown is in the prize pool section above.
How does the interview fast track work?
Every participant in our track is eligible for a fast-track interview for open roles at BasedAI. We are actively hiring, and what you build in a week is a better signal than any resume. We will reach out to participants we want to talk to after the event.
Where do I register?
On the official event page. The Apply button on this page links straight to it.
Bring your best week of building to the BasedAI track.
Imperial College London, 28 June – 4 July 2026. Real credits, real tooling, and a fast track into the team building the commercialization layer for open source AI.
Apply to the hackathon