Stop re-pasting project docs into every AI chat.

Most people are searching their first documents within minutes. Download the Windows installer, launch the app, create a namespace, and drag in a folder. On-device embeddings work out of the box — no Docker, Python environment, cloud account, or API keys required. Optional MCP setup for Cursor is one stdio command copied from Settings.

Aura Terran currently requires Windows. We recommend 32 GB RAM and an Intel Core i7 or equivalent processor. If you use Ollama for local models, an NVIDIA RTX 3090 or equivalent GPU is ideal. Cloud API providers do not require a local GPU.

Yes — RAG is how Aura Terran grounds agents in your private knowledge. Aura indexes and retrieves ranked chunks, and can optionally inject that context via LLM Passthrough before forwarding to your upstream model. Management API and inbound MCP remain available for chunk-only integrations. Passthrough also verifies replies against the injected chunks, and Conflicts / Retrieval Lab help keep the corpus and settings honest.

Both. Retrieval returns ranked chunks like any RAG stack. When you use LLM Passthrough, Aura sits in the request path and scores each claim in the model reply against the chunks it injected — grounded, weak, or unsupported. Grounding telemetry tracks which documents actually support answers over time, including never-used and high-traffic-but-never-useful sources.

Retrieval Lab has three tabs: Sample queries, Tune, and Baseline. Tune runs a saved question set through two retrieval configurations side by side and compares overlap, rank movement, latency, and better/worse verdicts on labelled questions. Experiments are per-search — they never touch saved settings, and live listeners keep serving traffic during a run. Baseline saves today's chunks per question and reports labelled regressions vs unlabelled drift after settings change.

Conflicts runs a local, deterministic sweep for passages that are largely the same text but state a different measurement, setting, or instruction. Automatic resolution settles disagreements the corpus already knows how to settle (version lineage, authority tiers, corroboration, recency) and leaves only undecidable fights for Prefer / Accept both. Settlements and your decisions are remembered so they survive re-ingest; the Activity log shows what ran unattended and lets you undo.

No. Free is free for evaluation. Paid editions are one-time purchases for the current major version. Future major upgrades may be paid, with discounts for existing paid customers.

Aura Terran Free lets you try the app locally with one knowledge base and up to 3 root folders (subfolders included). It is intended for personal evaluation and non-commercial use.

No. Commercial use requires Aura Terran Professional or Teams.

No for core search. Embeddings run on-device by default (ONNX Runtime), and hybrid retrieval works without any external model. OpenAI-compatible or Ollama endpoints are optional — for an ingest-time Chunking LLM under Settings → Chunking, optional Ollama embeddings, or LLM Passthrough upstream forwarding. At answer time you can also call rag_search / Management API and inject chunks into whatever chat model your client already uses.

In three places. (1) Optional Chunking LLM during ingest only — point Settings → Chunking at Ollama or an OpenAI-compatible API to refine split boundaries. (2) Optional LLM Passthrough at query time (Personal+) — inbound OpenAI and Ollama listeners retrieve, inject context, and forward to your upstream model. (3) Chunk-only path — clients get ranked chunks from Management API or MCP and inject them into any provider themselves.

Yes, optionally. Enable LLM Passthrough (Personal+) and point OpenAI- or Ollama-shaped clients at Aura's inbound listeners. Aura retrieves from your local corpus, injects context, then forwards to your configured upstream LLM. You can also skip passthrough and use Management API or inbound MCP for ranked chunks only — your client runs the completion.

Structure-aware splitting that respects OpenAPI specs, JSON, Markdown headings, and code boundaries — so chunks are not broken mid-function or mid-section. Hierarchical chunking adds coarse segments for broader context. Optionally enable a Chunking LLM (Ollama or OpenAI-compatible) under Settings → Chunking for finer paragraph seams on narrative documents; that runs at index time only. Rebuild the vector store after changing chunk strategy.

Aura Terran is a local-first desktop app. Your documents are indexed locally unless you explicitly configure remote model providers, web search, or external tools — those may send relevant request data to third parties.

A namespace is an isolated knowledge base or corpus. It lets you keep different projects or document sets separate.

Personal is for private individual use and unlocks multiple namespaces, Management API, OpenAI/Ollama passthrough, and automatic conflict settlement. Inbound MCP (IDE integration) is included on Free and every paid tier. Professional adds commercial solo use, external MCP connections, advanced retrieval, metadata filtering, web search, and server stats.

Teams adds per-seat or 5-seat-bundle licensing, priority support, deployment documentation, tool policy and approval rules, secret redaction, tool-pack import/export, and team-friendly controls on top of Professional's external MCP connections.

You keep the version you bought. Major upgrades may be optional paid upgrades, with discounted pricing for existing paid customers.

Setup and onboarding may be offered separately, especially for Teams and future Enterprise customers.

These are planned future Team and Enterprise capabilities. They are coming later and are not included in the initial paid checkout tiers.

Aura Terran indexes common document and code formats: PDF, Word, Excel, CSV, HTML, Markdown, plain text, and typical source and config files (JSON, YAML, and more). Add individual files, whole folders, or drag-and-drop from Explorer.

Enable the inbound MCP server in Settings and paste the stdio bridge command into your MCP client config. Agents call rag_search against your indexed corpus — the same hybrid retrieval pipeline as the built-in Search UI and Management API.

Yes. Personal, Professional, and Teams Terran purchases include the Aura Wire Terran RAG bridge entitlement at no extra charge. Activate the companion Wire key from your receipt (or merge it into an existing Wire Observe license) so Wire can inject ranked Terran chunks into chat. Terran Free does not include the bridge; Wire Professional also includes the same bridge entitlement.

Hybrid search combines keyword matching (BM25) with vector similarity so you get both exact-term hits and conceptually related chunks. You can also switch to pure vector, BM25, or TF-IDF modes in the UI or API when tuning retrieval.

Mark files or folders as Always Inject to prepend their chunks to every search result — style guides, security policies, or schemas that should always reach the model. Always-inject chunks do not count toward your Top K limit.

Yes for core indexing, on-device embeddings, and search. Paid licenses work offline when a valid stored license exists. An optional Chunking LLM needs network access if you point it at a cloud API; local Ollama chunking works offline. Optional web search blending and remote providers require network when configured.

Both hit the same retrieval engine and return ranked chunks — not chat completions. Management API (default port 41800) is for scripts, CI, and custom apps via POST /admin/corpus-search. Inbound MCP (default port 41801) exposes rag_search for IDE agents like Cursor and Claude Desktop via a stdio bridge.

Halo Softworks is a new software publisher, so Microsoft Defender SmartScreen may not yet have enough reputation data for every installer or release. Halo Softworks installers are digitally signed with Microsoft Azure Artifact Signing. Confirm that the publisher is Halo Softworks and that the digital signature is valid before continuing. The Secure Windows Downloads page on this site has step-by-step verification.

Yes. Halo Softworks signs Windows installers using Microsoft Azure Artifact Signing. You can verify the signature from the file's Properties dialog under Digital Signatures, or compare the SHA-512 checksum shown on the product download page.

No. Halo Softworks does not recommend disabling SmartScreen. Verify the publisher, digital signature, download domain, and SHA-512 checksum instead. See Secure Windows Downloads (linked from Download and the site footer).

Do not run the installer. Delete it and download a fresh copy directly from the Halo Softworks website. Contact support if the problem continues.