AI infrastructure. Memory. Reason. Act.
AIVAX brings models, retrieval and tools into one OpenAI-compatible operating layer, so behavior can evolve without losing the trail of what happened.
AIVAX / Since launch
Cumulative activity since AIVAX launched.
- Token volume
- —
- Token volume (USD)
- —
- RAG searches
- —
- RAG indexes
- —
Loading AIVAX platform activity.
AIVAX / Platform
Knowledge matters when its path into an answer remains visible.
Latest from AIVAX
View all posts
Inside one request
One request. A whole operating system behind it.
Keep the client simple. Let a gateway carry the instructions, knowledge, tools and policy that should be shared across applications.
See the request anatomy-
01 / Request
One familiar call leaves the application.
POST /v1/chat/completions -
02 / Gateway
Shared behavior takes shape behind it.
model · instructions · RAG · tools · moderation -
03 / Result
The answer returns with its trail intact.
choices[] · usage · generation_context
Everything the client doesn't have to carry.
Compose knowledge, model behavior and tools behind one OpenAI-compatible call. Each capability stays reusable, inspectable and shared.
Keep knowledge current, searchable and traceable.
RAG / 09 primitives
-
01
RAG grounding
Enrich a gateway with retrieved context while keeping the client unchanged.
-
02
Automatic reindexing
Regenerate indexed chunks and vectors when document content changes.
-
03
Bulk document import
Send larger document sets through a bounded import pipeline.
-
04
Document management
List, filter, update and remove documents and indexed chunks.
-
05
Collections over MCP
Expose collection search to agents as generated MCP tools.
-
06
Reranking
Refine retrieved candidates before they enter the model context.
-
07
Hybrid RAG (Reflex)
Combine semantic retrieval with deterministic lexical evidence.
-
08
Text classification
Assign text to defined labels through reusable model inference.
-
09
Text segmentation
Split long text into coherent sections for retrieval and processing.
Shape how context becomes a reliable answer.
Gateway capabilities
-
01
OpenAI-compatible inference
Route one familiar call across integrated and external models.
-
02
Subscription model reserve
Cover eligible model usage through plan-based reserve windows.
-
03
JSON Healing
Extract, validate and retry malformed JSON against your schema.
-
04
Structured responses
Use provider-native or AIVAX-managed JSON Schema output.
-
05
Skills
Load specialized system instructions only when a request needs them.
-
06
Multimodal preprocessing
Translate images, audio, video and files into text before inference.
-
07
Context control
Trim context and tool messages to fit the selected model.
-
08
Moderation
Apply input and output checks as reusable gateway policy.
-
09
Remote instructions
Load shared instructions from controlled remote sources.
-
10
Voice agents
Build conversational agents that listen, reason and respond in voice.
-
11
Text-to-speech
Turn generated text into natural audio through compatible models.
-
12
Speech-to-text
Transcribe spoken audio into text ready for models and workflows.
-
13
Image generation
Create images through the same managed inference layer.
Connect models and tools to systems outside the prompt.
Tools + providers / 08 primitives
-
01
Workers & hooks
Apply policy, enrichment and audit logic outside the prompt through authenticated HTTP hooks.
-
02
MCP & protocol tools
Attach server-side tools to the same request path.
-
03
Tool-call interpreter
Let models without native tool calling invoke functions.
-
04
Shell sandbox
Run shell tools inside isolated sandboxed environments.
-
05
BYOK
Connect an external OpenAI-compatible provider through the gateway.
-
06
Web search
Give agents current web results through a managed search tool.
-
07
OCR
Extract machine-readable text from images and scanned documents.
-
08
Web fetch
Retrieve and normalize page content for agents and pipelines.
A shared path matters when every application can rely on it.
Field guide
Five ways into the system.
Start with the question in front of you: knowledge, execution, evaluation, model choice or cost.