Configure. Ground. Govern.
Knowledge assistants you can operate on your own terms.
OrcAI helps individuals, specialist teams, educators, and research groups build AI assistants around curated knowledge bases. Upload material, shape assistant behaviour, control access, and run the stack without depending on external SaaS services beyond an inference provider.
Grounded chat
People ask questions in focused chats backed by selected source material instead of relying on generic model context.
Reusable assistants
Teams compose bots from purpose, response behaviour, retrieval repositories, sharing rules, and publication status.
Content library
Upload documents, Office files, PDFs, images, and metadata-backed source items for reuse across bots and chats.
Retrieval and citations
Processed material is indexed for semantic retrieval so answers can refer back to the knowledge base behind them.
Controlled access
Organizations manage groups, invitations, visibility, direct grants, and capability-gated administration.
Self-hosted stack
Run the full application, workers, storage, vector search, authorization, and databases under your own control.
Working model
From source material to governed assistant.
Generic chat tools make people improvise context, policy, and quality checks. OrcAI turns those concerns into explicit workspace resources: content, repository blocks, behaviour blocks, bots, access settings, providers, and quotas.
Add source material to a governed content library.
Workers extract, process, and index content for retrieval.
Package material into repository blocks with retrieval settings.
Build assistants from behaviour rules, repositories, model choices, and access policy.
Users chat with assistants that retrieve relevant context from the selected knowledge base.
Admins manage providers, models, quotas, groups, sharing, and operations.
Governance
Control is part of the application, not an afterthought.
OrcAI is built for settings where the knowledge base, access model, model configuration, and operating environment matter. The stack is open source and can be hosted with infrastructure you control.
Authorization you can reason about
Organization context, capability-gated UI, groups, invitations, visibility settings, and resource-level grants are built into the core workflows.
Provider and quota control
Admins can configure model providers, available models, usage quotas, and the operational limits that shape AI access.
Retrieval as infrastructure
Content ingestion, object storage, workers, embeddings, and Qdrant indexes make knowledge bases reusable rather than ad hoc chat attachments.
Self-hosting
A full stack you can inspect.
The application runs with its own app process, background workers, relational data, cache, object storage, vector search, and authorization service. The required external boundary is an OpenAI-compatible inference endpoint for embeddings and model calls.
Self-hosting docsSpecialist teams
Turn internal documents, procedures, and domain references into assistants for repeatable knowledge work.
Individuals
Work against a curated personal or project knowledge base without manually pasting context into every prompt.
Education
Instructors can still create course-specific assistants and learners can ask against selected material.
Administrators
Keep model providers, quotas, groups, resource access, and operational dependencies explicit.
Research groups
Experiment with retrieval-augmented workflows in a structured platform whose moving parts can be inspected.
Self-hosters
Operate OrcAI with your own PostgreSQL, Valkey, S3-compatible storage, Qdrant, SpiceDB, and inference endpoint.
Use cases
Useful wherever knowledge needs context and boundaries.
Course and seminar assistants grounded in selected teaching material.
Project knowledge bases for teams working through technical, legal, policy, or research documents.
Internal procedure assistants that answer from approved operating manuals and reference material.
Research prototypes for evaluating retrieval-augmented chat in controlled organizational settings.
Context and constraints
Open source application stack designed to run without external SaaS dependencies beyond an inference provider.
Self-hosting is supported with Docker Compose or manually managed services.
Governance is part of the product model: organizations, groups, grants, providers, models, quotas, and workers are first-class concerns.
Answer quality depends on the selected model, embedding configuration, indexed material, and retrieval setup.
OrcAI is in active development and not yet presented as a production-hardened operations guide.
Developed in the Sokratesᵗ and KI:edu.nrw applied project context at Rhine-Waal University.