Ground your AI in governed, approved knowledge

AI agents answer questions with cited sources, and approved data sets publish incrementally to your vector stores, RAG and private LLM applications.

Request a demo Talk to sales

Key features

  • AI agents that cite their sources

    Answers point back to the node, license and changeset they came from.

  • Publish to RAG apps

    Approved changes reach your vector databases as incremental data sets.

  • Private models

    Run Gitana and your models on your own network, or mix hosted and private.

AI agents that cite their sources

Agents take on repetitive editorial work and answer questions about your content, always pointing back to the node and changeset they used.

  • Helpful assistants

    Agents do their work in isolated branches and hand it back to you for editing and approval, so nothing ships unreviewed.

  • Conversational insights with citations

    Ask about one document or a whole collection. Every answer cites where in your content it came from, including the rights that apply.

  • Content generation

    Agents use hosted or private LLMs to summarize documents, draft questions and answers, extract terms and produce key points.

To an AI agent approved data set · master

Can we use the Hamlet summary in our EU catalog?

No. Example Press licenses the summary for non-commercial use only, which excludes a catalog. The EU is a cleared territory, so the license is the only blocker.

Hamletstore:book
License
Editorial, non-commercial
Territories
US, EU
Changeset
44:95f85022
Release
Spring 2026
Example answer after the Spring 2026 merge, grounded in an approved node, with its license and changeset. Illustrative data.

Publish approved content to RAG apps

Your editorial teams keep the content behind retrieval-augmented generation (RAG) and custom LLM applications current, and only approved changes reach them.

  • Data sets from editorial changes

    Server-side scripts turn changed content into data set rows for your vector database, built incrementally from each change.

  • Test, preview and validate

    Deploy a data set to staging first and score it against your RAG app or fine-tuning endpoint before it goes live.

  • Scheduled publishing

    Releases deploy incremental data sets to production vector databases or LLM endpoints, with the embedding model and vector size you choose.

master → vector DBPublishing flow
  1. Approve on masterbranch: masterEditors merge reviewed changes to the approved branch.
  2. Generate the data setincremental deltaA script maps the changed nodes and their relationships into JSON or JSONL rows.
  3. Preview in stagingvector DB: stagingThe delta is deployed to staging and scored against expected answers.
  4. Publish on schedulevector DB: productionThe release sends the same delta to production with your embedding model.
How an approved change reaches a RAG application.

Private models, and data that stays on your network

Run Gitana and your LLMs on premises, in the cloud or as a hybrid, and connect any mix of public and private models and vector databases.

PrivateGPT and local LLMs
Deliver trained data sets to local LLM or PrivateGPT endpoints for incremental fine-tuning.
Hosted model fine-tuning
Send data set deltas to fine-tuning endpoints for hosted models such as OpenAI, or for models you run yourself.
Keep it local
Run Gitana and your models on the same private network so source content never leaves it.
Self-managed Gitana
Access control and audit
Access policies decide who and what can read content. On dedicated Gitana Cloud and self-managed installations, the audit engine records every access.
Auditing

Data sets, training and continuous fine-tuning

Train and validate models on approved content, alongside the branches and releases your editors already use.

Automatic data sets
Background jobs build training and validation sets from your content graph as it changes, kept separate so models can be tuned and then scored.
Training pipelines
Pre-train a model on content scheduled for a release and swap it in at launch, or train a branch-specific model for in-app A/B tests. Training runs as a distributed job.
Scoring and iteration
Validation runs report which tests failed and an overall score. Adjust content and graph weights, retrain and compare scores.

Pre-built integrations with Google Vertex AI, Azure Machine Learning, AWS, OpenAI and IBM Watson, or connect your own pipelines.

See how Gitana fits your content

In a demo, a solutions engineer walks you through Gitana around your use case. The same call covers architecture, security, deployment and procurement questions. In a guided evaluation, you work on your own content with one of our engineers on Gitana Cloud.

Request a demo Request a guided evaluation