A modular AI platform that lets companies use their own accumulated data to build “small language models” (SLMs) specialized for a business domain, and deploy them in the cloud, at the edge, or on-premises. Founded in Seattle in 2023 by former Microsoft platform engineers, it offered an alternative — “build small, and own it yourself” — for workloads where general-purpose large language models do not add up in terms of accuracy, cost, or data handling. In November 2025, its acquisition by Cisco was announced and completed, and its capabilities are being folded into Cisco AI Canvas, the company’s enterprise AI workspace. The standalone official site does not respond as of August 2026, so anyone considering it now should check how it is offered on the Cisco side.
Key Features
- Building SLMs from your own data: Internal documents, historical records, and operational logs are used for training to create a language model narrowed to a specific business domain. Rather than having a general-purpose model answer broadly and shallowly, the design aims for a model that answers reliably within a narrow scope
- Modular end-to-end foundation: Provides a set of modules covering data ingestion, model training, evaluation, and deployment as a single flow, so you can combine only the stages you need
- Choice of cloud, edge, or on-premises deployment: In addition to SaaS use, it supports placement in your own data center or on equipment at the site (the edge), leaving room for adoption in industries that cannot move data outside
- Model weight ownership and data sovereignty: It emphasizes that the enterprise can hold the weights of the trained model, which suits organizations that want to limit dependence on external services
- Continuous learning and compliance monitoring in operation: Includes a mechanism to keep updating the model from data patterns observed in real-world use, along with monitoring from a regulatory-compliance perspective
- Lower cost through smaller models: Compared with a setup that constantly calls a huge general-purpose model, it positions reduced inference cost and computing resource requirements as its main appeal
Pricing
| Plan | Price | Main contents |
|---|---|---|
| Enterprise (standalone offering before the acquisition) | Contact for pricing | A full foundation for SLM building, training, and deployment. Individually quoted; no public price list was published |
| Via Cisco AI Canvas (current offering) | Included with an eligible Cisco license | Offered as part of Cisco’s enterprise AI workspace. Detailed terms need to be confirmed with Cisco |
Pricing is as of August 2026. The standalone NeuralFabric site does not currently respond, so please check the Cisco AI Canvas official page for the latest terms.
Pros & Cons
✅ Pros
- Because the model is small and limited to a business domain, inference cost and computing resources are easier to keep down than with a general-purpose model
- Since your own data is used for training, it is easier to get responses that assume internal terminology and your own operating procedures
- Supports on-premises and edge placement, so it can be considered even in industries that cannot entrust data to outside parties
- Owning the model weights in-house reduces exposure to a vendor’s policy changes or service shutdowns
- Everything from data ingestion to operation is handled on one foundation, reducing the patchwork of multiple tools
⚠️ Cons
- It assumes you already have enough in-house data of sufficient quality; organizations whose data is not in order will see little benefit
- An SLM deliberately covers a narrow range, so general-purpose answers on unexpected topics cannot be expected
- No public pricing is shown, making it hard to estimate the cost of a deployment in advance
- With the Cisco acquisition, integration into Cisco’s product lineup is advancing and the path to adopting it as a standalone platform is unclear
- The standalone official site does not respond as of August 2026, and primary information has moved to the Cisco side
Comparison with Similar Services
| Item | NeuralFabric | Cisco AI Canvas | NVIDIA NeMo | Databricks Mosaic AI |
|---|---|---|---|---|
| Provider | Cisco (via acquisition) | Cisco | NVIDIA | Databricks |
| Main use | Building and deploying SLMs from your own data | AI workspace for IT operations | Training and customizing your own models | Model building and operation integrated with a data platform |
| Target model size | Small (domain-specific) | Use of purpose-built models | Small to large | Small to large |
| Deployment targets | Cloud / edge / on-premises | Integrated with Cisco products | Cloud / on-premises | Mainly cloud (on the data platform) |
| Intended users | Enterprises with proprietary data | IT operations teams | Organizations running their own AI infrastructure | Companies already operating a data platform |
Who Is It For
- Companies for which general-purpose AI services are not accurate enough and that need responses aligned with their own business vocabulary
- Organizations whose regulations or security policies prevent data from leaving the company, making on-premises or edge operation a requirement
- Teams whose API charges for large models have piled up and who want to rethink the structure of their inference costs
- Enterprises that want to hold model weights as a company asset and build a setup that is not at a vendor’s mercy
- IT departments that already use Cisco products and are considering an in-house AI configuration combined with AI Canvas
Summary
NeuralFabric packaged a direction — instead of leaving everything to a general-purpose large model, build a small model from your own data and own it yourself — into an enterprise-grade foundation. It is a practical answer to the three enterprise AI challenges of cost, data sovereignty, and accuracy, and that way of thinking has not lost its validity as of 2026.
At the same time, the acquisition by Cisco completed in November 2025, and the capabilities are being integrated into Cisco AI Canvas. Since the standalone official site no longer responds, anyone considering adoption should realistically check how it will be delivered within Cisco’s product line. If the “own your SLM” design philosophy appeals to you, it is worth comparing it side by side with other platforms taking a similar approach.