> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ragen.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Ragen AI Roadmap: Current and Planned Upcoming Features
> Explore what Ragen AI is actively building right now, what is planned for the next milestone, what comes later, and what is out of scope.
Ragen is actively developed, and this page shows the shape of what is coming — both what is being built right now and what is planned further out. Nothing here carries a release date. Open work lives in [GitHub Issues](https://github.com/webamigos/RagenAI/issues), and this page changes when a real install needs something that wasn't anticipated.
## Choosing Your Own Infrastructure
Three parts of Ragen are things you should be able to pick rather than inherit: what runs background jobs, what stores vectors, and what parses documents. Today each is a single choice baked into the stack. The next set of changes puts a seam in front of each one and adds a second implementation behind it.
Nothing is being taken away. Temporal stays a supported runtime, Qdrant stays the default vector store, and Docling stays the default parser. If you change nothing, nothing changes.
| Concern | Today | Also Available | What the Alternative Costs You |
| ---------------- | --------------------- | ------------------------------------------- | -------------------------------------------------------- |
| Background jobs | Temporal | BullMQ on Redis | A crashed job re-runs from the start instead of resuming |
| Vector store | Qdrant | `pgvector`, in the database you already run | A shared failure domain, and different retrieval numbers |
| Document parsing | Docling (self-hosted) | Mistral Document AI (hosted) | **Your documents leave your deployment** |
Two things worth understanding before assembling the smallest possible install:
* **The hosted parser is a data-processing decision, not a performance one.** With `DOCUMENT_PARSER=mistral`, every ingested document is transmitted to a third party. That is the right trade for some deployments and disqualifying for others, which is why it will never be a silent fallback. See the [Security](/security/overview) page.
* **`pgvector` is a different retrieval system, not a port of the current one.** Qdrant scores sparse vectors server-side; Postgres has no equivalent, so the lexical half of retrieval changes and the results change with it. It ships with published measurements or it does not ship.
The designs are public, one file each, in [`docs/specs/`](https://github.com/webamigos/RagenAI/tree/main/docs/specs) — they name the trade-offs and failure modes rather than only the upside.
## Also Next
These are the things actively in progress alongside the infrastructure flexibility work above.
The TypeScript SDK is official and published. Python is the language most people integrating the API are actually writing in, and a Python client is next.
An assistant you talk to in a channel, carrying the same retrieval-time access filter as the main application. This is distinct from the Slack connector that reads Slack as a source.
A local model, local embeddings, no outbound path, and a script that demonstrates it end to end. The pieces exist today — what's missing is a single verified configuration.
SSO and MFA are not built yet. Entra ID over OAuth comes first, with SAML and SCIM to follow after.
A `docker compose up` command that starts Ragen with sample documents and sample questions already loaded, so evaluating Ragen doesn't begin with an empty knowledge base.
SharePoint, OneDrive, Outlook, and Teams connectors to bring your Microsoft 365 content into a Ragen knowledge base.
In-application feedback on answers — the signal that retrieval quality work has been missing.
## Later
These are planned but not actively in progress. They are on the list because real installs will eventually need them.
* **Permissions inherited from the source system**, so revocation in Google Drive or SharePoint takes effect without waiting for a re-index.
* Confluence and Jira connectors.
* Web search and multi-step research as opt-in tools, because some installs deliberately have no outbound path and that must remain possible.
* Audit log and security-event export to a SIEM, with retention and automatic deletion.
* Organization export, so leaving Ragen is a documented procedure.
* Fail-closed encryption: refuse to start in production without a key provider, rather than starting and storing message content unencrypted.
* MFA and passkeys.
* Per-user permissions for individual MCP tools.
* Usage drill-down from organization to team to user to request, with budget alerts.
* A sizing guide, a versioning policy, and a CHANGELOG.
## Not Planned
Worth saying plainly, because it saves you evaluating Ragen for something it is not going to be.
* **Image generation, advanced voice mode, a code interpreter.** Ragen answers from your documents. A general-purpose assistant is a different product, and there are good ones.
* **A knowledge graph layer.** Improving retrieval that can be measured is the priority over adding a stage that cannot.
## Follow Open Work
All open issues are tracked publicly. If something you need isn't listed here, open an issue — this roadmap changes when a real install needs something unexpected.
Browse open issues, follow progress on active work, or file a new request.