OpenInterpretability
  • Ekbasis
  • Guide
  • Research
  • Lab
  • Tools
  • Notes
  • Registry
  • Manifesto
Download the modelDownload
OpenInterpretability

Ekbasis: know what an action will do before your agent runs it. From an independent lab for AI-agent safety. Open weights, Apache-2.0: download it and run it yourself.

Download the model
Ekbasis
  • Getting started
  • Download (Hugging Face)
  • For agents (agents.md)
  • Results and models
  • Client and CLI
  • Cookbook
The lab
  • About the lab
  • Open tools
  • Observatory
  • ProbeBench
  • InterpScore
  • Academy
Research
  • Manifesto
  • Roadmap
  • Papers & posts
  • Docs
Community
  • GitHub org
  • Notebooks
  • SDK (openinterp)
  • HuggingFace
© 2026 OpenInterpretability — Apache-2.0 for code, CC-BY 4.0 for docs.Built in public.
Back homeOPEN SOURCE · APACHE-2.0

Tools

What came out of the research, released so anyone can reproduce it, extend it, or use it to keep agents safe.

The agent safety stack

Four questions before an agent acts; each layer sees what the others cannot.

  1. 01 · provenance & policy

    Who wants this action?

    Does it derive from untrusted data, and are its parameters within policy? Dataflow, not text, so obfuscated injections are caught.

    AgentGuard L0–L1 →

  2. 02 · intent

    Does the agent mean harm?

    A late-layer direction reads whether the agent is committed to an unauthorized irreversible action. Needs open weights.

    AgentGuard L2 →

  3. 03 · consequence

    What will it do here?

    A world model trained on real executions predicts what the action will do in the current state, calibrated, in one pass. Works with closed agents.

    Ekbasis →

  4. 04 · actuation

    What happens now?

    Block, redirect to a safe read-only action, or escalate to a human, with the reason each layer gave.

    AgentGuard L3 →

Agent safety

EkbasisNew

An open world model for agents: what an action will do in this state (lose work? fail?), calibrated, one pass. Git guard, Claude Code hook, MCP server.

AgentGuard

Defense-in-depth action firewall for tool-using agents, with a model-internal intent brake.

FabricationGuard

Activation-probe fabrication detection for open-weights LLMs: AUROC 0.88 cross-task, ~1 ms. pip install openinterp.

agent-probe-guard

Two-probe activation gate for code agents: predicts whether a trace will succeed before tools fire.

Research instruments

openinterp-mcp

Bring-your-own-agent interpretability: probe-causality experiments from Claude Code or Cursor.

openinterp-lab

One-command replication of the papers on the Google Colab CLI.

decision-locator

Find the layer where a language model commits a decision — and steer it.

inspect-tool-entropy-collapse

Inspect AI eval that detects the WANDERING failure mode in agent trajectories.

openinterp-swebench-harness

Instrumented agent harness that records feature trajectories during SWE-bench runs.

Benchmarks & standards

ProbeBench

Leaderboard and registry for activation probes, with error analyses.

InterpScore

Composite, transparent ranking of sparse autoencoders.

Training

mechreward

Interpretable internal features as dense reward signals for RL training of LLMs.

notebooks

Train your first sparse autoencoder in 30 minutes, then scale to 27B.

openinterp (SDK + CLI)

Python SDK and command line for the probes and the leaderboard.