Request a quote

Tell us how you want to run kiLM

kiLM ships in three deployment profiles and runs on either GPU or non-GPU hardware. Pricing depends on the profile, the data domains you turn on, the connectors you opt into, and your scale. Fill in what you know — anything you skip we'll cover in the follow-up call. A member of our team replies within one business day.

Quote request form

Every field stays on this page until you submit. We never send form contents to analytics or ad networks, and don't store anything until you click "Submit request." If your network blocks our portal, use the email fallback at the bottom.

Step 1

Engagement stage

Where are you in your evaluation? This drives the pricing tier, the support SLA, and whether we issue a short-lived trial license or a production license.

PoC Proof-of-concept on a sample corpus. Free, time-boxed (30 days), 1–2 evaluators. Hosted by us or by you depending on data sensitivity. Trial Time-boxed trial on your real data, in your environment. 60–90 days, full feature set, capped concurrency. Converts to Production on signed order. Beta Limited production with a defined user cohort. Used when a feature is shipping but not yet at GA SLA. Discounted price; production support level. Production Full GA deployment with production SLA, full support tier, annual license. This is the path most customers take after PoC or Trial.

Step 2

Deployment model

Who runs the cluster? Managed service has us operating the control plane in a single-tenant VPC; on-premise has you operating it in your own infrastructure.

Managed service We host the control plane and dispatch patches via your AWS / Azure / GCP account. Single-tenant per install; isolated VPC; least-privileged IAM. No air-gap option (managed implies connectivity to us for patches + telemetry). On-premise You operate kiLM in your own infrastructure — your data center, your VPC, or your private cloud. Updates arrive as signed bundles you apply on your terms. Next step: pick air-gap or connected. Not sure yet We'll walk you through the trade-offs in the follow-up call. Most regulated customers start with on-premise air-gap; Managed suits faster time-to-value.

Step 4

Outgoing-intelligence policy

Should kiLM ever send your data OUT of the cluster — e.g. cloud-LLM prompts to OpenAI / Anthropic / Vertex, webhook deliveries, or vendor telemetry? Intake-only is the default for sensitive deployments; "Outbound allowed" unlocks cloud-LLM components and outbound integrations in the next step.

Intake-only kiLM pulls FROM your approved sources but never emits OUT. Cloud-LLM components are hidden in the next step. Default for sensitive deployments. Air-gap implies intake-only by definition. Outbound allowed kiLM may send data to allow-listed outbound destinations. Required for cloud LLMs (OpenAI, Anthropic, Vertex, Bedrock, Azure), webhook delivery, and vendor telemetry. Not compatible with air-gap.

Step 5

Hardware shape

kiLM runs on CPU-only hardware. GPU is required only for large local LLM inference, vision encoding, and fine-tuning workloads — the rest of the platform (ingestion, extraction, embedding, retrieval, graph) runs comfortably on CPU.

CPU-only (non-GPU) Suitable for ingestion, hybrid retrieval, graph, and remote-LLM modes. Lowest hardware cost; lowest LLM throughput unless you bring an external endpoint. GPU-accelerated Required for fully-local LLM, fully-local vision encoding, and fine-tuning. We size to your throughput; NVIDIA L40S / H100 / A100 classes supported. Hybrid (CPU + remote LLM) Run everything except LLM inference locally; route prompts to your existing internal LLM endpoint. Common for customers with central AI platforms.

Hardware envelope (GPU details)

Which GPU tier?

kiLM's runtime LLM router (slice 0200) picks the most-capable model that fits your hardware. Pick the tier that matches your install — we'll seed your install's hardware profile so the router starts making the right call from day one.

Consumer GPU (8–16 GB VRAM) e.g. RTX 3060, RTX 4070, M2 Pro/Max. Runs llava + small quantized models. Cannot run 70B-class even at q4. Workstation GPU (24–48 GB VRAM) e.g. RTX 4090, RTX A6000, A100-40. Runs Mixtral 8x7B q4. Adequate for most enterprise workloads. Datacenter GPU (80 GB single) e.g. H100 80GB, A100 80GB. Runs Qwen 2.5 72B, DeepSeek-Coder 67B, and Llama 4 Scout at q4. Multi-GPU (160+ GB total) e.g. 2× H100, 4× A100. Required for frontier MoE models > 200B total weights or fp16 70B+.

GPU count Total VRAM (GB across all GPUs) GPU model (optional)

We use these values to seed your install's hardware profile so the LLM router picks the right model immediately. You can change them later via /admin/install-hardware.

Step 6

Scale and topology

Rough numbers are fine — we use these to scope the cluster size and the license tier.

Topology Expected named users

Expected document volume Monthly net-new ingest

Step 7

Your industry

Helps us bring the right reference architecture and content packs to the call. Pick the primary one.

Industry

Step 9

Existing data infrastructure

kiLM can either ingest from your warehouse / lake in place (no byte copy) or pull-and-stage. Knowing what you already run helps us scope the right mode upfront.

Data warehouse / SQL platform Data lake / object storage (system of record)

Approximate lake size today Preferred warehouse mode (if any)

Interested in lake federation (kiLM stores the index but not the bytes — recommended if the lake is > 10 TB).

Step 10

Region & compliance

Tell us where the install runs and any frameworks your security team will sample evidence against. Everything below is optional but speeds up the security review.

Primary region Data residency requirement

Frameworks your security team will sample against (select all that apply)

SOC 2 (Common Criteria) HIPAA Security Rule GDPR / EU privacy ISO 27001 NIST 800-53 CMMC / DFARS ITAR / EAR controlled Japan APPI Other (note below)

Step 11

Licence & pricing

Pick the licence tier, the capability bundles you need, and the capacity tier. The price is computed live against our published rate card. Managed substrate adds a per-bundle infra uplift (we pay the AWS bill); Customer- cloud and Air-gap don't — you pay your own infrastructure direct.

Licence tier

Trial Evaluation only. Ollama-only LLM, no outbound emit, no GPU, low monthly caps. Cost-contained by design. Beta Early-access programme. Subsidised add-ons. Suitable for production pilots up to mid-volume. Production Full enterprise licence. All bundles purchasable. Capacity tiers from Standard to Enterprise.

Capability bundles

Greyed checkboxes are not available for your selected substrate + tier combination (e.g. outgoing-intelligence is hard-blocked on Air-gap). Hover for the bundle's gate list.

Loading capability matrix…

Individual components (optional)

Pick exactly the per-vendor integrations and AI substrate pieces you need (each priced individually). Categories expand on click; selecting a component adds its monthly fee to the live total below. Leave blank to take only the bundles above.

Loading component catalog…

Capacity tier

Standard 8 vCPU / 32 GB RAM / 200 GB storage envelope. Suits departmental pilots and small-team production. Pro 32 vCPU / 128 GB / 1 TB. 1h critical SLA on Managed. Suits most production deployments. Enterprise 128 vCPU / 512 GB / 5 TB. 99.95% uptime, DR drills, dedicated CSM on Managed.

Estimated monthly pricing

Itemised below. Total is binding only after vendor review (margin- cover check) and your formal acceptance. All figures USD.

Pick a tier + bundles + capacity to see your estimate.
Total / month	—

Step 12

Who should we reply to?

We send the quote to the work email below. Add a buying-team contact if procurement should be looped in from the start.

Full name Work email

Company Role / title

Phone (optional) Country

Target start Indicative annual budget

Step 13

Anything else?

Specific use cases, must-have connectors, PoC criteria, RFP windows, the auditor you need to satisfy — anything that helps us send you a sharper quote.

Notes We require a mutual NDA before substantive discussions. I have read the privacy policy and consent to Polycracy processing the information above to respond to my quote request.

Network blocking the portal? Send it by email instead — opens your mail client with the answers pre-filled.

We don't publish list-price configurators on this site. Pricing depends on the deployment profile, the connectors and data domains you enable, your scale, and your support commitments. The quote we send back is specific to this form — not an auto-generated number from a marketing calculator.