GAIA [RESEARCH]GitHub ↗

MISSION CONTROL / OPEN RESEARCH COLLECTIVE

Pushing the Limits of AI Agent Capabilities

Gaia Research is the open research collective defining, verifying, and publishing the frontier of AI agent skills.

Explore research →View the ledger →
Chief Scout mascot in the Gaia Research laboratory

CHIEF SCOUT / SIGNAL ACQUIRED

01flagship labContext Diet
04benchmark tracksIn formation
05ledger recordsPublicly scoped

Static launch indicators — live registry sync is pending.

01 / PROJECT GATEWAYS

Choose your entry point.

Discovery in the lab. Permanent records in the tree.

01

Gaia Skill Tree

The canonical registry for named, durable capabilities.

Enter gateway
04

Community

Bring the test, the counterexample, or the next fusion.

Enter gateway
05

Docs and specs

Schemas, templates, and implementation notes—without gatekeeping.

Enter gateway
02 / PUBLIC EVIDENCE

Research Ledger

Every capability claim gets a traceable state. Live registry synchronization is not connected yet.

Read the protocol →
Selected Gaia Research ledger records
RecordTypeStateEvidence
Context Diet / Lab 001ProtocolACT○ Source pending
G-Bench Generalist SuiteBenchmarkPRP○ Specification pending
Skill Evidence SchemaSpecVRF↗ Repository schema
Memory Architecture AnalysisResearch briefREV○ Review queue
Protocol Fusion NotesPaperACT○ Source pending
03 / TEST RANGE

Benchmark Suites

Compact, legible evaluation surfaces. Data collection is staged; labels show the work honestly.

01 / Generalist suite

G-Bench

Capability coverage across tools, planning, and execution.

Scope defined · results pending
02 / Eval set

Safety alignment

Tests for refusal quality, constraints, and recovery.

Scope defined · results pending
03 / Architecture analysis

Memory

What persists, what decays, and what stays attributable.

Scope defined · results pending
04 / Active research

Protocol fusion

How compatible skills compose under explicit evidence.

Scope defined · results pending

04 / OPEN SCIENCE PROTOCOL

Evidence moves. Nothing disappears.

Research owns discovery and verification. Gaia Skill Tree owns inscription and permanence.

  1. 01Observe
  2. 02Benchmark
  3. 03Verify
  4. 04Publish
  5. 05Register

05 / SECONDARY BRIDGE

Gaia Skill Tree remembers.

When a capability survives the lab, it earns a durable place in the canonical registry.

Visit Gaia Skill Tree →
HUNTER'S ATLAS / PREVIEWVerified capabilities
deserve a permanent trace.Registry destination · external
Milim, Chief Capability Scout

06 / CHIEF SCOUT CONSOLE

Milim Directives

  1. DIR-01Share your research. Open science is the way.
  2. DIR-02Verify before you flex. Data beats hype.
  3. DIR-03Fuse ideas. Break silos.
  4. DIR-04Build safe, robust agents.

07 / OPEN INVITATION

Bring proof. Build the frontier.

The lab is public by design. Contribute a result, challenge a claim, or help shape the next benchmark.

Contribute on GitHub ↗Read the Codex ↗