What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Basegent’s support assistant for Bucket Space used Sanity Context MCP as a read-only bridge to support content, while Basegent handled the assistant and retrieval workflow and Bucket Space displayed the chat. In the project author’s account, an important lesson was that an outline alone did not always route a question to the right entry: richer index metadata and search helped the implementation find relevant material. This is a reported build, not an independently tested performance study.
How the Basegent and Bucket Space support setup worked
In a case study published October 4, 2026, author Miracleio describes Bucket Space, a bookmarking and content-capture application, as a proving ground for Basegent, an AI customer-support and operations platform. The implementation combined three parts:
- Sanity: hosted support content in a Knowledge Base, including policy and setup information.
- Basegent: connected to that content through a first-party Sanity Context source adapter and managed the assistant and retrieval orchestration.
- Bucket Space: embedded the Basegent chat widget in the application.
The separation matters: Sanity Context exposes content; it does not run the model or agent loop. Sanity describes Context as a hosted MCP server that gives AI agents structured, read-only access to content. It does not write back to the dataset. Sanity Context documentation
In the author’s project, Knowledge Base entries covered matters such as refund rules by plan, market, and product or deployment type, plus setup guides such as an iOS Share Sheet shortcut. The author says issue review surfaced conflicting policy details, which the team resolved into standing instructions. Those examples describe this project’s content and editorial workflow, not a guarantee that Sanity will identify or resolve policy conflicts for every Knowledge Base.
#1 Best Overall
How the assistant retrieved support information
The reported retrieval flow began by fetching a Knowledge Base outline, selecting relevant paths, and reading entries over MCP. Sanity’s documented Knowledge Base tools include initial_context, which returns the outline, and knowledge_base_read, which retrieves entry content using a Knowledge Base ID and path. Sanity’s tool reference says to use paths verbatim from the outline. Sanity Context MCP tools
The author also reports using knowledge_base_search to help find candidate paths. Treat that as a detail of the Basegent adapter described in the case study; check Sanity’s current tool reference for supported tools and behavior when implementing your own connection.
Rank #2
Why an outline may not be enough
Miracleio recounts an outline-routing miss when a customer asked how to set up an iOS shortcut. The implementation was then changed to add richer index metadata and native search alongside outline-based routing. The practical lesson is that a navigable outline may not describe every entry in the terms customers use. A retrieval system can benefit from search signals and metadata that connect customer phrasing—such as “bookmark links from Safari”—to the relevant setup guide.
This is a project lesson, not a measured success-rate improvement. The case study does not publish a controlled retrieval benchmark or quantify how often the revised approach found the right entry.
Choosing between live GROQ access and a Knowledge Base
Sanity documents two Context modes. Choose based on how your content is organized and when the assistant needs to see changes.
| Approach | Content and timing | Agent access | Important qualification |
|---|---|---|---|
| GROQ mode | Queries a configured live dataset; suited to structured, schema-consistent content. | Schema and query tools. | Access is limited by the configured dataset and permissions. |
| Knowledge Base mode | Retrieves material selected and indexed ahead of time; useful when relevant information is distributed across prose or sources. | Outline and entry-reading tools, including initial_context and knowledge_base_read. |
Knowledge Bases are opt-in beta; features and limits may change before general availability. |
These distinctions follow Sanity’s Context overview and Knowledge Bases documentation. The Knowledge Bases page was last updated September 18, 2026 and identifies the feature as opt-in beta. Confirm current availability and limits before building around it.
What to plan before connecting an agent
Use the right token and keep it server-side
The Context MCP endpoint is organization-scoped. Sanity’s reference gives the endpoint shape as https://api.sanity.io/v1/context/organizations/:organizationId/mcp/:mcpEndpointName. Requests use a bearer token, and the documented requirement is an organization API token with Context Viewer permission; a project API token is not accepted. Keep the credential on the server rather than exposing it in a browser-based chat widget. See Sanity’s Context MCP reference.
Design the application layer around the content layer
Because Context is read-only and does not run the agent loop, the application still needs to decide how to select and present retrieved material, handle user identity, and respond when retrieval is incomplete. The author reports passing authenticated user attributes with support requests, distinguishing refund eligibility from processing time in an example answer, and offering an escalation route. These are implementation choices in the case study, not built-in Context guarantees.
Best Value
Plan for imperfect routing and response failures
The article also describes handling a repeated escalation prompt and a stream that yielded no tokens. These details illustrate operational cases an application may need to address; they do not establish general reliability or provide a tested error-handling recipe. Decide how your assistant should recover when it finds no relevant entry, produces an empty response, or cannot resolve a policy question, and make escalation available where a wrong answer could have consequences.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the case study does—and does not—show
Miracleio reports using a multi-provider BYOK/BYOI arrangement and published @basegent/react and @basegent/client packages. The case study is an account of one implementation, not an independent audit. It supplies no controlled benchmark, quantified support-resolution improvement, uptime result, or independent security review. It therefore supports a description of the architecture and lessons the author reports, not a claim that the setup eliminates hallucinations or improves support outcomes by a known amount.
For a similar project, the durable design idea is to keep the knowledge source, retrieval/application logic, and embedded interface as distinct layers. Then validate the paths customers actually use to ask questions, rather than assuming a tidy outline will always match their wording.
Recommended Free Tools
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
