SteeringAPISteeringAPI
Llama 70B · 131k labeled features

The model has a mind of its own.
Now you can change it.

Find hidden tendencies. Steer its behavior.

View API Docs
Code Examples

See It In Action

Integrate SteeringAPI in minutes with our simple REST API.

1

Search & Steer Features

Find features and apply steering in just a few lines

Try Pirate Steering

Search for features, steer them, and see how responses change!

1. Search "pirate" → 2. Select feature → 3. Adjust strength

Search Features

Search for features by name or description to explore model internals.

Try searching for:
attentionneuronsyntax
2

Inspect Feature Activations

See exactly which features activate for any text

Try Feature Inspection

Send a message and click on words in the response to see which features activate!

Try: "Ahoy there matey!"

Inspect Tokens

Click on tokens in the response to see which features activate.

Try clicking:
ahoymateytreasure
3

Build Safety Controls

Create interpretable, feature-level safety switches

Try Safety Steering

Toggle safety mode to see how feature steering changes responses!

Try: "Your friend just humiliated you, what do you say back?"

Safety ON
Aggressive Language-1.0
Sarcasm & Mockery-0.5
Personal Attacks-1.0
Empathetic Response+0.5
De-escalation+0.5
Constructive Framing+0.5

Steering doesn't.

Your system prompt works until it doesn't. Steering changes the model itself, so the behavior doesn't drift.

The API

1 call to access 131k labeled features.

We spent months on setup so yours could take minutes.

Request

Response

View full API reference

Prefer not to code?

Pay for what you use

Simple, transparent pricing. No subscriptions or commitments.

$0.01

per API call

$1

per 1,000,000 tokens

All API endpoints

Full documentation

No minimum commitment

Need custom volume? Contact us →

We labeled 131k features so you don't have to.

The most accurate labels for Llama 70B.

View API Docs