Agentic RAG
Best-in-class retrieval quality with hybrid keyword + vector search and AI agents that plan, retrieve, and reason.
A self-hostable enterprise AI platform for retrieval-augmented generation, deep research, custom agents, and agentic workflows — over every system your team uses.
From hybrid retrieval to agentic research, voice, and code execution — AetherSearch unifies the toolbox for grounded, auditable AI in your stack.
Best-in-class retrieval quality with hybrid keyword + vector search and AI agents that plan, retrieve, and reason.
Multi-step research flow that generates rigorously cited, in-depth reports — built on iterative agentic exploration.
Build agents with bespoke instructions, tools, knowledge, and actions — composable for any workflow.
Pluggable web providers — Serper, Google PSE, Brave, SearXNG — for fresh, citable answers from the open internet.
Generate documents, graphics, and downloadable assets directly from a chat — output that goes beyond text.
Let agents talk back to your tools — read and write across systems with flexible auth and a Model Context Protocol bridge.
Run code in a sandbox to analyze data, render charts, transform files, and explore datasets in-conversation.
Speech-to-text and text-to-speech for natural conversation, plus prompt-driven image generation.
Every layer is replaceable, observable, and self-hostable. Nothing leaves your perimeter unless you say so.
Documents flow through a streamlined ingestion pipeline coordinated by Celery. Every step is idempotent, observable, and resumable.
AetherSearch separates concerns across thread-pool workers so heavy tasks never block the request path. Each worker is multi-tenant, observable, and independently scalable.
Coordinates connector lifecycle, vespa sync, pruning checks, and system-wide periodic tasks.
Pulls documents from external sources, batches them, and spawns docprocessing tasks. Watchdog monitors stuck connectors.
Upserts to Postgres, chunks, contextualizes, embeds via model server, and writes to Vespa.
Vespa metadata sync, doc permission upserts, checkpoint cleanup — fast tasks at high throughput.
Connector pruning, document permissions sync, external group sync, large CSV exports.
Builds the Knowledge Graph by extracting relationships between documents and running clustering.
Tracks queue depth, process memory, system health, and Celery liveness across the cluster.
Indexes user-uploaded files and synchronizes them into the project workspace they belong to.
The DynamicTenantScheduler dispatches periodic tasks: indexing checks, sync probes, KG rounds, cleanups.
Hook into the systems your team already uses. AetherSearch keeps every document, ticket, and message indexed, permission-aware, and continuously fresh.
Pragmatic technology choices: production frameworks, mature data systems, and a pluggable AI layer.
Next.js 15 · React 18 · TypeScript · Tailwind · Radix · Headless UI · dnd-kit · Storybook
Python 3.11 · FastAPI · SQLAlchemy · Alembic · Pydantic · Celery · LangChain
PostgreSQL (multi-tenant) · Vespa hybrid index · Redis · MinIO/S3 blob store
LiteLLM bridge · Anthropic · OpenAI · Gemini · Ollama · vLLM · custom embeddings & rerankers
OAuth2 · OIDC · SAML · SCIM · RBAC · encrypted credentials · audit log
Pick the footprint that fits the moment. Scale up when you're ready, never lose data ownership.
Sub-1GB memory footprint. Get a chat UI & agents running in minutes.
The complete stack with hybrid index, workers, and indexing pipelines.
Add governance, identity, analytics, and whitelabeling for org-wide adoption.
AetherSearch is open source, MIT licensed, and built for the long run.