| What it does | Citations, built for AI agents
| On-device moderation that flags hateful, abusive and threatening text in 23 European languages, before the message is sent.
|
|---|
| Stage | Not given
| MVP
|
|---|
| Founded | Not given
| Not given
|
|---|
| Based in | Not given
| NL
|
|---|
| Pricing model | Freemium
| Freemium
|
|---|
| Pricing | Free tier for small requests with no API key required; 10 million free tokens to start; $1 per million tokens for pay-as-you-go, with discounts available at volume or with annual commitments; failed requests are never charged.
| Free up to 100k monthly active devices with unlimited inference. Custom licenses available.
|
|---|
| Key features | - Token-level attribution with confidence scores
- Works with any AI model without access to weights or internals
- Processes documents up to 1 million tokens
- Returns results in under 2 seconds for 20,000-token documents
- On-premise and air-gapped deployment options
- PDF support with character-level mapping
- Supports 23+ languages including Indian languages
- Chrome browser extension for page and PDF tracing
- Raw token-by-token heatmaps available via API
- No data logging or training on request content
| - Three separate scores: hateful, abusive, threat
- Ten target signals under the hate score: race, color, religion, descent, national or ethnic origin, sexual orientation, gender, disability, age, other
- Runs fully on device with no server calls
- Supports 23 European languages from one model file
- 80MB model size optimized for iOS, Android, and browser
- 0.847 macro F1 accuracy on Multilingual HateCheck
- Warn before message is sent
- Comment triage without infrastructure
|
|---|
| What makes it different | - Measures attention directly from the model's computation rather than asking the model to cite itself, providing verifiable attribution
- Achieves frontier-grade citation quality at 7x lower cost and 5-6x faster than alternatives
- Never confused by duplicate facts—identifies the exact occurrence the answer used even when values appear multiple times
- Lands on exact tokens down to character level, not just document chunks
- Works on any model's output post-generation without modifying inference pipeline
| - Multilingual support across 23 European languages in a single 80MB model
- Complete privacy with no data leaving the device
- Free access up to 100k monthly active devices with unlimited inference
|
|---|
| Community votes | 0 | 0 |
|---|
| DR (Domain Rating by Ahrefs) | 2 ● no change since last check | 37 ● no change since last check |
|---|
| Trust Flow (Majestic) | 3 | 20 |
|---|
| Citation Flow (Majestic) | 16 | 39 |
|---|
| Referring domains (Majestic) | 36 | 616 |
|---|
TokenPath
Toxic