---
title: Enterprises are building private, curated AI — not buying another chatbot
description: LangChain, AlphaSense, Wren AI and Oracle all made the same pitch this week in different words: keep the knowledge and the data inside your own walls, and let agents work from there.
canonical: https://hangingcontext.com/stories/2026-11-22/enterprises-want-ai-on-their-terms/
last-updated: 2026-11-22
url: https://hangingcontext.com/stories/2026-11-22/enterprises-want-ai-on-their-terms/
published: 2026-11-22T00:50:00Z
edition: 2026-11-22T00
newsroom: artificial-intelligence
beat: developers
trust: confirmed
source_count: 4
cited_manifest_ids: ['1795286518430907277', '1795225713427703056', '1795285563622899371', '1795237356007743378']
---

# Enterprises are building private, curated AI — not buying another chatbot

*LangChain, AlphaSense, Wren AI and Oracle all made the same pitch this week in different words: keep the knowledge and the data inside your own walls, and let agents work from there.*

Confirmed · 4 Manifests · Artificial Intelligence · Published 2026-11-22 00:50 UTC

## Key facts

- LangChain CEO Harrison Chase says the missing piece for enterprise AI is better knowledge, not better models, and proposes a knowledge factory with a curator agent that maintains it in the background. (self-reported) [cites: 1795286518430907277]
- AlphaSense says its connectors link over 20 applications and combine internal signals with 500 million external documents, with answers linking back to sources and following existing permissions. (self-reported) [cites: 1795225713427703056]
- Wren AI and TrueNAS say their private GenBI setup can run fully air-gapped with no cloud LLM in the loop, querying data in place with SQL written by open-weight models on local GPUs. (self-reported) [cites: 1795285563622899371]
- Oracle proposes a healthcare reference design on OCI using AMD SEV-SNP and COCONUT-SVSM, where a customer-operated verifier checks boot evidence and the key broker withholds the data key if verification fails; the attestation services are experimental. (self-reported) [cites: 1795237356007743378]

## The knowledge factory

LangChain CEO Harrison Chase put a name on the problem this week: the knowledge factory. His argument is that the missing piece for making AI work inside a company is not better models and not just more context, but better knowledge. A knowledge factory is a system that collects, curates and maintains a representation of company knowledge to pass to agents. Knowledge, in his framing, is the reusable understanding that lets an agent do a task correctly: what a metric means, which source to trust, how a particular situation is handled. That is different from context, which is what you hand an agent for one specific task. [cites: 1795286518430907277]

Chase divides the factory's knowledge into three kinds: skills, meaning procedures; domain knowledge, meaning policies and facts; and guidance on data access. The maintenance is meant to be automatic, not manual. A curator agent runs in the background, on some cadence, updating knowledge from raw sources and agent traces. Agents themselves should be given a tool to flag knowledge that is missing or wrong, and to suggest the correction. The goal is a system where doing the work reveals what knowledge is missing, and fixing those gaps makes future work better. Chase is candid about the state of the field: there is no industry-standard format for semantic memory in AI agents yet, so he advises against being too prescriptive about the format. LangChain says its own internal go-to-market and content studio agents are powered by a knowledge factory. [cites: 1795286518430907277]

## Research, centralised

AlphaSense makes the same argument from the research side. The company says its platform grounds AI-generated deliverables in a firm's own knowledge and market data, by integrating internal business systems with its SuperAnalyst AI. Its connectors link over 20 applications, and the company says it combines live signals from those systems with 500 million external documents. Answers link back to their underlying sources, and access follows the permissions already set in each system, which is the part that matters for a deliverable a firm has to defend. [cites: 1795225713427703056]

> **From Synorb** — Save what you follow as a Beacon. A Beacon is a saved query that your agent pulls with an explicit lookback window. Every new account starts with 5 starter Beacons: Business & Technology Podcasts, AI Companies, Venture Capital, Private Companies and Public Company Filings & Executive Transitions. [See Beacons](https://synorb.com/connect/start)


The plumbing is deliberately uneven. SharePoint, Intralinks and Gong connectors check for changes every 15 minutes, while Slack and Salesforce connectors query the source in real time. Bipsync notes sync with their labels as searchable metadata, and the Gong connector syncs transcripts rather than audio. AlphaSense says it is used by 92 percent of the S&P 100, 95 percent of the top consulting firms, 84 percent of the top asset management firms and 90 percent of the top investment banks. Those are the company's own figures, but they point to the same pattern: research AI is being sold as a centralised system, not a search box. [cites: 1795225713427703056]

## Ask the data where it lives

Wren AI and TrueNAS take the private version of the argument all the way to air-gapped. At SC26 in Chicago this week, TrueNAS joined Phison and Wren AI as a hardware partner for private generative business intelligence. Wren AI sits inside the network, next to the data on TrueNAS, and the companies say there is no cloud LLM in the loop and the deployment can run fully air-gapped. Databases, files and object data stay on TrueNAS and are queried in place, with no ETL and no migration. [cites: 1795285563622899371]

The SQL is written by a validated open-weight model on the customer's own GPUs, served by an endpoint the customer runs such as Ollama or vLLM, or delivered on the Phison AI Data Platform. Row- and column-level security, validated SQL and a full audit trail apply before any query runs. Questions can come from a person in Slack or Teams, or from an AI agent over MCP. TrueNAS is the storage and data-protection layer, with end-to-end checksums, snapshots and replication, and Wren AI stores its context models, metric definitions and audit logs on TrueNAS. The pitch is aimed squarely at teams that cannot let data leave the building. [cites: 1795285563622899371]

## The guardrails are the product

Oracle's contribution this week is the most careful of the four, and the most honest about its limits. Its architects propose a reference design for protecting healthcare pipelines on OCI, covering clinical trial analytics, genomic analysis and electronic health records, which need protection during processing as well as evidence that the processing environment started in an approved state. The design targets a customer-managed SEV-SNP guest on supported OCI E5 and E6 bare metal. AMD SEV-SNP encrypts confidential virtual machine memory and adds integrity protections against malicious memory remapping, and the memory-encryption keys are held in the AMD Secure Processor, unavailable to the hypervisor, OCI and other tenants. [cites: 1795237356007743378]

The trust mechanism is a customer-operated verifier that evaluates boot evidence; if verification fails, the key broker withholds the healthcare data key. COCONUT-SVSM provides isolated services inside the confidential virtual machine, including its virtual Trusted Platform Module at VM privilege level 0. But Oracle is explicit that this is a proposed reference design, not a standard OCI-managed feature, and that it needs specific integration and testing. The COCONUT-SVSM attestation services are described as experimental, with key-authentication and rollback limitations. Verification and key-release audit records support governance reviews, the company says, but they do not automatically establish regulatory compliance. Shielded Instances and Confidential Computing cannot be enabled together on OCI. Read together, the four announcements describe one shift: enterprises are not asking for a smarter chatbot, but for a system that knows what the company knows, keeps the data where it is, and can show its work. [cites: 1795237356007743378]

## What is still unresolved

- Whether the knowledge factory idea converges on a standard format for semantic memory; LangChain says none has emerged yet.
- How many enterprises will accept the integration work Oracle's healthcare reference design requires, given it is not a standard OCI-managed feature.
- Whether air-gapped GenBI deployments like the Wren AI and TrueNAS setup perform well enough on local open-weight models for everyday business questions.

## Decision card

- **Useful for:** AI executives planning enterprise agent rollouts; Teams choosing between hosted AI and private deployments; Healthcare and research leaders handling sensitive data
- **Not for:** Teams that want a quick chatbot over public data; Anyone without someone to own the knowledge the agents rely on
- **Compare against:** A hosted copilot or chatbot that sends company data to a vendor cloud; A retrieval setup that indexes documents but never curates them

## Provenance ledger

| Manifest id | Title | Source | Claims used | Published |
|---|---|---|---|---|
| `1795286518430907277` | How to build a knowledge factory | [langchain-blog](https://langchain.com/blog/knowledge-factory) | 1795286782773909449, 1795286782776909450, 1795286782781909455, 1795286782777909451, 1795286782778909452, 1795286782780909454, 1795286782783909457 | 11-21 00:00 |
| `1795225713427703056` | How to Centralize Your Research and Get Better Deliverables | [alphasense-product-articles](https://alpha-sense.com/resources/product-articles/centralize-research) | 1795225951196704994, 1795225951202704995, 1795225951231704999, 1795225951189704993, 1795225951214704997, 1795225951225704998, 1795225951237705000, 1795225951239705001 | 11-20 00:00 |
| `1795285563622899371` | TrueNAS × Wren AI: Private GenBI on Storage You Already Trust | [wren-ai-blog](https://getwren.ai/post/wren-ai-truenas-private-genbi) | 1795285809418899934, 1795285809436899939, 1795285809442899940, 1795285809402899932, 1795285809410899933, 1795285809432899936, 1795285809434899938, 1795285809426899935 | 11-21 00:00 |
| `1795237356007743378` | Helping Protect Healthcare Pipelines from Boot to Memory on OCI | [oracle-blog](https://blogs.oracle.com/cloud-infrastructure/protect-boot-to-memory-on-oci) | 1795237625618744058, 1795237625624744065, 1795237625624744064, 1795237625617744057, 1795237625621744061, 1795237625622744062, 1795237625619744059, 1795237625625744066, 1795237625623744063 | 11-20 00:00 |

## Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

1. **Get a Synorb Key.** Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool `synorb-get-demo-key` (no header needed), or REST `POST https://api.synorb.com/api/connect/demo-key` with a JSON body such as `{"client":"your-agent-name"}`. The key is the `api_key` field of the answer. Save the key. Existing 72 Hour Agent Demo keys keep their issued limits. New keys require email OTP verification. An existing demo key lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons.
2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer <your Synorb Key>`. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.

Cited Manifest ids:

- `1795286518430907277`
- `1795225713427703056`
- `1795285563622899371`
- `1795237356007743378`

```sh
for id in 1795286518430907277 1795225713427703056 1795285563622899371 1795237356007743378; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done
```

> **From Synorb** — Claims with a source link. A Manifest holds a Signal of structured claims and a Brief, plus the source it came from. Your agent can check a sentence against it. [Read the guide](https://synorb.com/connect/start)

## Related stories

- [Legal adopted AI fastest. Now it has to prove the money was well spent](https://hangingcontext.com/stories/2026-11-21/legal-ai-cant-prove-it/)
- [Agents need trusted metadata, so governance is moving into the data layer](https://hangingcontext.com/stories/2026-11-21/data-governance-ai-control-plane/)
- [AI is starting to do the research, and the tests can barely keep up](https://hangingcontext.com/stories/2026-11-21/ai-researching-itself/)
- [Health systems scale AI by standardising first: the boring lesson behind the savings](https://hangingcontext.com/stories/2026-11-20/health-systems-standardize-before-automating/)

## Links

- HTML: https://hangingcontext.com/stories/2026-11-22/enterprises-want-ai-on-their-terms/
- JSON: https://hangingcontext.com/stories/2026-11-22/enterprises-want-ai-on-their-terms/index.json
- Markdown: https://hangingcontext.com/stories/2026-11-22/enterprises-want-ai-on-their-terms/index.md

Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.

---
Reuse (terms v1): Free to read, quote, summarise and republish, including commercially. Credit Hanging Context and link to the story. Keep the cited Manifest ids when you republish claims. Do not imply that Synorb endorses you or your product. Credit: "Hanging Context, with provenance by Synorb", with a link to https://hangingcontext.com/stories/2026-11-22/enterprises-want-ai-on-their-terms/. Terms: https://hangingcontext.com/terms/reuse/
