Every major model, one future endpoint
Compare models across providers by modality, context length, and indicative price today. Knowledge bases, guardrails, and an OpenAI-compatible gateway are on the way.
12
Providers
From OpenAI to Hugging Face
Weekly
Usage ranking
Sorted by tokens used last week
3
Guardrail types
PII, jailbreak, moderation
5
Embedding models
For knowledge bases
What flowbit AI is
flowbit AI is the organization's home for working with models, available today as a preview. Its goal is one OpenAI-compatible HTTP contract behind flowbit — chat, embeddings and images first — so the upstream provider can change without changing your application.
The Model Marketplace is live: a catalog of models from twelve providers — including OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, xAI, Cohere, NVIDIA, Z.ai, Hugging Face and Together AI — ranked by last week's usage and refreshed hourly. You can filter by provider, modality and context length and compare indicative prices per million tokens.
Knowledge Bases and Guardrails are interactive configurators. A knowledge base combines one of five embedding models with a managed PostgreSQL plan for vector storage; a guardrail sets sensitive-data protection, jailbreak protection or content moderation, with mask, block or log-only modes for PII. The Memory Playground shows how an assistant could organize what it learns about a user.
Because this is a preview, inference, API keys, document ingest, guardrail enforcement and billing are not enabled yet, and every price is an indicative USD figure. The layout and contracts are designed now so you can plan against them.
Models from
- OpenAI
- Anthropic
- Meta
- Mistral
- DeepSeek
- xAI
- Cohere
- NVIDIA
- Z.ai (GLM)
- Hugging Face
- Together AI
Four surfaces, one mental model
Model Marketplace
Search and filter by provider, modality, and context length. Sort by most used, recommended, name, or newest.
Knowledge Bases
A RAG configurator: embedding model, a PostgreSQL plan, and a live specification summary.
Guardrails
A two-step policy wizard for sensitive-data protection with mask, block, or log-only modes.
Memory Playground
An interactive memory map that shows how an assistant could organize what it learns about a user.
OpenAI-compatible, provider-replaceable
Apps will call stable HTTP endpoints behind flowbit. The upstream provider — and the GPUs underneath — can change without touching your code.
- Chat, embeddings, and images first; agent runs later
- One flowbit key instead of a key per vendor
- Not live yet — shown to document the target API
curl "$FLOWBIT_AI/v1/chat" \
-H "Authorization: Bearer $FLOWBIT_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-sonnet-5",
"messages": [
{ "role": "user", "content": "Summarize this changelog." }
]
}'Embedding models in the configurator
| Model | Provider | Dimensions | Context | Input / 1M tokens |
|---|---|---|---|---|
| text-embedding-3-large | OpenAI | 3072 | 8,191 | $0.13 |
| text-embedding-3-small | OpenAI | 1536 | 8,191 | $0.02 |
| gemini-embedding-001 | 3072 | 2,048 | $0.15 | |
| embed-v4.0 | Cohere | 1024 | 128,000 | $0.12 |
| bge-m3 | Together AI | 1024 | 8,192 | $0.008 |
Indicative USD pricing, preview only.
Vector storage on managed PostgreSQL
pg-starter
$12/ month · $0.017/h
- vCPU1
- Memory1 GB
- Storage10 GB
- Regioneu-central-1
pg-standard
$48/ month · $0.066/h
- vCPU2
- Memory4 GB
- Storage50 GB
- Regioneu-central-1
Safety policies, configured before they are enforced
| Guardrail | What it does | Indicative / 1M tokens |
|---|---|---|
| Sensitive data protection | Detect and redact emails, phone numbers, payment cards, government IDs, and API keys — mask, block, or log only | $0.35 |
| Jailbreak protection | Flag prompt-injection and jailbreak attempts | $0.25 |
| Content moderation | Score and filter harmful content in inputs and outputs | $0.20 |
Common questions
Can I call models through flowbit today?
Not yet. The marketplace is a live catalog, but inference, API keys, and billing are not enabled in this preview.
Where does the catalog come from?
A public, usage-ranked model list refreshed hourly, mapped onto flowbit providers, with a curated fallback if the fetch fails.
Do knowledge bases ingest my documents?
Not in the preview. You can configure a knowledge base end to end; Create stays disabled until ingest ships.
Are the prices final?
No. All prices are indicative USD figures for comparison.
Explore the Model Marketplace
Compare models from 12 providers in one place.


