InterpretabilityMapping latent goals in open-weight agents
A reproducible probe of emergent planning circuits across three model families.
11 min read
The AGI Scientist is a community-driven platform for frontier research — open experiments, interoperable tools, and the people turning ideas into projects that ship.
Every breakthrough begins in the open. Post a hypothesis, frame a problem, and invite the community to sharpen it with you.
Pin your environment, data, and code. Results others can re-run are results others can build on.
Graduate experiments into interoperable tools and projects that anyone can plug into their own stack.
Ship it in the open, get peer-reviewed, earn credit — and hand the next researcher a higher starting point.
From the first question to a shipped tool — everything is open, interoperable, and built with the community.
Peer-shared experiments, reproducible benchmarks, and living papers — published in the open and built to be challenged.
Learn moreCollaborate on ambitious AGI projects. Fork, contribute, and ship — with credit and governance built in.
Learn moreA curated, searchable registry of agents, models, and utilities. Vetted, versioned, and ready to plug in.
Learn moreOpen standards and adapters so tools, datasets, and agents speak the same language across the stack.
Learn moreDiscussion, mentorship, and working groups. Find collaborators and turn ideas into funded projects.
Learn moreGuides, primers, and curated reading paths — from first principles to frontier research.
Learn moreReproducible, peer-reviewed, and free to read. Publish your findings and let the community build on them.
InterpretabilityA reproducible probe of emergent planning circuits across three model families.
11 min read
BenchmarksWhy current evals miss the point, and an open protocol that doesn't.
8 min read
AlignmentDesigning incentives so autonomous researchers share, not hoard.
14 min readFork, contribute, and ship. Governance and credit are built in, so collaboration stays fair.
A self-hosting autonomous research environment that runs and documents its own experiments.
Shared schema + adapters letting any agent discover and call any compliant tool.
One-click reproduction of published results, pinned to exact environments and data.
Vetted, versioned tools that speak a common protocol. Discover, plug in, and compose.
Compose multi-agent pipelines visually.
Reusable, sharable benchmark harness.
Version, clean, and trace datasets.
Inspect activations in any model.
Policy-as-code for autonomous systems.
Sandboxed execution for agent code.
Thousands of scientists, engineers, and builders — across labs, startups, and bedrooms — collaborating in the open. Find your people and your next project.
Self-organizing teams around interpretability, evals, agents, and safety.
Get feedback from people who've shipped — and give it back.
Pitch in the open, find collaborators, and graduate into incubation.

Open research stays free, forever. Membership funds the compute, the events, and the people — and unlocks deeper collaboration.
For the curious and the newcomers.
For active builders and contributors.
For teams, labs, and sponsors.
One-time gifts and sponsorships keep open research open.
Join researchers and builders shaping the open path to AGI. No spam — just signal, every week.
12,000+ members · Free to join · Unsubscribe anytime