AiGrid is in beta, with trials scoped to your estate.Request a briefing

Beta

Questions, answered directly

If yours is not here, ask it in a briefing and we will answer it with the product in front of you.

What it is

What is AiGrid?

AiGrid is a self-hosted appliance that makes your organization's documents searchable by your AI assistant. It uses compute you already own to read, OCR, and index the documents inside your network, and serves the result through one retrieval endpoint. Every result is a passage from your own documents with its filename and page number.

Does AiGrid generate answers?

No. AiGrid retrieves and never generates. The endpoint returns the passages that best match a question, and your AI assistant reads them and writes the answer with citations. No model in the product writes answers or summaries, so the product itself has no way to invent one.

Is AiGrid a cloud service?

No. There is no AiGrid cloud service and no AiGrid account. The appliance is a container image that you run on your own single-node Kubernetes or Kubernetes cluster, and the documents, extracted text, index, and models all stay in your environment.

What is the release status?

AiGrid is in beta and is running with beta customers. Evaluations are run as a commercial trial that we scope with you during a briefing.

Compute

Where does the compute come from?

From capacity you already own and are not using: workstations after hours, servers in your DR datacenter, and the excess capacity in your dev, test, and production clusters. You group machines into pools and set the working hours for each pool.

Does AiGrid need GPUs?

No. Layout analysis and OCR run on ordinary CPUs across your pools, and the embedding models run on the appliance. All of the models are small enough to run on ordinary processors. If a machine in a pool has a GPU, the worker uses it, and otherwise it runs on the CPU.

Which machines can join a pool?

Windows 11 desktops, using the container support built into Windows under WSL, and Kubernetes clusters with spare capacity. Each machine needs at least 4 GiB of memory made available to the worker.

How is AiGrid installed on our workstations?

You install the AiGrid agent on the workstations you want to onboard, using Microsoft Intune, Group Policy, or whichever software deployment tool you already use. Once it is running, every connection the machine makes is outbound, and it only works inside the hours set for its pool.

How is AiGrid installed on our Kubernetes clusters?

You deploy the AiGrid agent as a manifest, using whichever deployment tool you prefer: Portainer, Argo CD, Flux, or kubectl. The agent runs below every other workload on the cluster and gives capacity back when the cluster needs it.

Do we have to use workstations?

No. A pool can be made up of servers only, such as a DR site or spare capacity on your Kubernetes clusters. Some organizations start there and add workstations once their desktop and security teams are comfortable.

Will staff notice AiGrid on their desktops?

It is designed to stay out of the way. An administrator sets the schedule for each pool, so workstations can be limited to out of hours. The agent runs at low priority, so it does not compete with other work, and on a desktop it runs under its own service account and not as the user.

What happens if a machine is switched off mid-document?

Nothing is lost. Machines that sleep, crash, or get unplugged are expected. The machine's claim on the document lapses and the document returns to the queue for another machine to pick up.

How long does it take to index our documents?

It depends on the size of your pools and the hours you give them, so we do not quote a rate. The dashboard reports progress as pages converted and passages indexed. Search stays available while documents are still being ingested.

Data and security

Do our documents leave our network?

No. Documents stay in your own S3-compatible storage, and AiGrid is only ever given read access to it. Processing and search call no external service, and all models ship inside the product images.

Does AiGrid copy or move our documents?

No. Your documents stay where they are, and AiGrid only ever has read access to them. A machine fetches one document at a time to convert it, using temporary storage while it works. What the appliance keeps is the extracted passages, each pointing back to its source file and page.

Where does the index live? Is it spread across the machines in the pool?

No. The machines in a pool only convert documents. The index and vector database sit on the appliance, which runs on your own Kubernetes, or in your own Elasticsearch-compatible database if you choose to use one.

How do we keep HR or finance documents separate from everything else?

Today one search token reads everything that has been indexed, so results are not filtered by who is asking. Index only what every user of that endpoint is allowed to see, or run a separate appliance for each group whose documents must stay apart.

Do machines in the pool receive inbound connections?

No. Every connection is outbound HTTPS from the machine to the appliance and to your document store. No port is opened on any desktop.

Can AiGrid run in an air-gapped environment?

Yes. Once installed, document processing and search need no internet access, and the documentation covers how to carry out the installation in an environment without internet access.

How do operators sign in?

Through your own identity provider using OpenID Connect, for example Entra ID or Okta, with a local administrator account as a fallback.

Using it

Which AI assistants can use AiGrid?

Any assistant or agent that can call an HTTPS endpoint. The dashboard provides a small download with connection instructions and the appliance's certificate. Claude-based tools are the documented example, and the endpoint itself is agent-agnostic.

Which document sources and file types are supported?

AiGrid reads from S3-compatible object storage that you own, and can index several buckets on one store. It converts PDF, Word (DOCX), Excel (XLSX), PowerPoint (PPTX), RTF, HTML, Markdown, AsciiDoc, and CSV files, along with pictures that contain text (PNG, JPEG, TIFF, BMP, and WebP), which it reads with OCR. For pictures it extracts the text in the image and does not interpret what the image shows. New and changed documents are picked up automatically.

Our documents are in a cloud bucket. Does AiGrid still apply?

Yes. AiGrid reads from S3-compatible object storage, and that can be your own object store or a bucket with a cloud provider. The conversion and indexing still run on machines you own.

What happens when we move a file?

Passages are linked to the file's location, so moving a file causes that file to be ingested again from its new location.

We already run our own RAG pipeline. Can we still use AiGrid?

Yes. You can move the whole RAG layer outside the appliance and keep AiGrid purely as a text extraction tool. It does the heavy document conversion on your idle compute and delivers clean markdown, with page markers, to a bucket you own. Your pipeline does the rest.

Can we use our own vector database?

Yes. The appliance includes a built-in vector database, and you can point it at one you already run instead. Tested with Elasticsearch 9.5. Other Elastic-compatible databases should work. You can also go further and move the whole RAG layer outside the appliance, keeping AiGrid for text extraction only.

We have already vectorized our documents. Is AiGrid still useful?

It can be. New and changed documents keep arriving, and AiGrid picks them up automatically. If you want to keep the pipeline you have, AiGrid can run purely as a text extraction tool, using your idle compute for the expensive conversion step and handing clean markdown to your existing pipeline.

Pricing and trial

How is AiGrid priced?

An annual fee for the appliance, with support included, plus an annual fee for each machine that joins the processing pool: one rate per Windows desktop and one rate per server (Kubernetes node). The pricing page has the current figures and an estimator.

Do you charge per user, per query, or per page?

No. The price is the appliance and the machines in the pool. The number of people or assistants searching, the number of queries, and the number of pages indexed do not change it.

How do we estimate the return?

Start with your page count. Cloud extraction services bill per page, so their cost grows with the size of your archive, while AiGrid is priced on the appliance and the machines you lend. The Compare page works through an example at published cloud list prices, including the point below which a cloud service costs less.

Is there a free trial?

Not as a self-serve download, because standing AiGrid up meaningfully involves enrolling real machines. It is evaluated under a commercial trial that we scope with you, and the evaluation runs on your infrastructure.

Ask us the hard ones

Request a briefing