# The Research Workbench: How the Tools Connect

URL: https://research-memex.org/docs/implementation/research-workbench
Description: Follow a research workflow from selected sources to a checked note. See what moves between tools, what you check and what to set up for your task.

*See where each tool sits before you install any of them.*

The research workbench helps you turn selected sources into notes you can check and use. Start with the task, then choose which parts need tools. You can follow the example below without installing anything.

Your library and notes can stay in files you own. That does not mean all processing stays on your computer. A host sends the selected text to the model you choose, whether you paste it or a connected tool reads it. Check what you may share before sending material to a service.

This page draws the configuration the systematic-review module ran. Cherry Studio is the host because most participants arrive from a chatbot, and it lets them compare models side by side before an agent touches their files ([Tools & budget](/docs/case-studies/systematic-reviews/systematic-review-syllabus#tools--budget)). Claude Code or another agent host can take the same position, with the same connections beneath it.

```text title="diagram:research-workbench"
+---------------------------------------------+
| Researcher                                  |
| question / judgment / checks                |
+----------------------+----------------------+
                       |
                       v
+----------------------+----------------------+
| Host: Cherry Studio                         |
| chat / agents / knowledge base              |
+---------+-----------------------+-----------+
          ^                       ^
          |                       |
          v                       v
+---------+----------+  +---------+----------+
| Tools through MCP  |  | Models through API |
| Zotero MCP         |  | gateways:          |
| filesystem MCP     |  |   OpenRouter       |
|   (vault only)     |  |   Haimaker         |
| Research Memex     |  | local model        |
|   MCP server       |  |   (optional)       |
+---------+----------+  +--------------------+
          ^
          |
          v
+---------+----------+  +--------------------+
| Durable record     |  | Discovery          |
| Zotero library     |<-+ Research Rabbit    |
| Obsidian vault     |  | OpenAlex           |
| papers as Markdown |  +--------------------+
+--------------------+
```

Read this connected configuration from the top. You direct the host, which calls tools and models and receives their results. The model sees the text supplied through the host, not unrestricted access to your computer. A scoped tool can save a result to your records. On the manual route, you supply the passage and save the checked note yourself. Discovery tools suggest sources; you decide which ones to keep.

## A round trip [#a-round-trip]

Follow a question from source selection to a checked research note. This is a constructed teaching example, separate from the course exercises. The source, draft and correction below are written for this demonstration. They are not a published study, live search or recorded model run.

Each row below lights one part of the drawing above: Discovery, then the Durable record, then the Host with its Tools and Models, then the Durable record again. The Researcher box is active in every row; that column is "What you check".

Question: does checking AI-generated briefs improve team decisions? Keep sources that describe teams checking such briefs. Distinguish evidence of finding errors from evidence of better decisions.

| Step | What moves | What you check |
|---|---|---|
| Find candidates | In a real search, terms such as "AI briefs" and "team decisions" go to OpenAlex or a discovery tool. Its candidate records go into a search log. Here, the supplied source is S01 below. | Does each source exist, and does it fit the question? A candidate list is not a complete systematic search. |
| Keep a source | A selected paper and its bibliographic record go into Zotero. For this demonstration, S01 is a local teaching text, not a paper to cite. | Can you return to the original passage? Keep its source ID and locator with the text. |
| Read and draft | You read S01, or pass its exact text and the question to an approved host. A connected host may retrieve selected material through Zotero or filesystem access. | May this text be sent to that model? Is the draft limited to what the passage supports? |
| Check and save | Compare the draft with S01, then save the revised note and your decision in a Markdown file. Obsidian can display that file. | Can another reader follow the locator and see why you revised the claim? |

### Inspect the supplied source [#inspect-the-supplied-source]

Constructed source S01, "Brief-checking pilot":

- S01.P1: In a four-week pilot, one team checked an AI-generated brief against its source documents before each planning meeting.
- S01.P2: The team recorded three unsupported statements and revised the briefs. The pilot did not measure decision quality or compare teams.

Draft claim: "Checking AI briefs improves team decisions (S01.P1-P2)."

The locator is present, but the claim goes beyond the source. S01.P2 reports corrections, not better decisions. A citation alone does not make a claim sound.

### Save the checked note [#save-the-checked-note]

You could save this demonstration note as `notes/brief-check.md`:

```markdown
# Brief-checking pilot

Source: S01, constructed teaching text, paragraphs P1-P2.
Observation: one team recorded three unsupported statements and revised its briefs.
Limit: decision quality and differences between teams were not measured.
Decision: keep as an example of error checking, not evidence of better decisions.
Next question: what evidence would show a change in decision quality?
```

On the manual route, you read, revise and save it. With connected tools, the host can retrieve the approved source and write a draft to an allowed folder. You still check the passage and decide whether to accept, revise or defer the claim. Do not overwrite the source or your independent first reading.

## Check your understanding [#check-your-understanding]

Use these as ungraded self-checks before adding tools. For assigned work, follow your instructor's brief; this example does not replace it.

- [ ] I can name the question, selected input and expected output.
- [ ] I can say where the source and checked note live.
- [ ] I can use S01.P2 to explain why the draft claim needed revision.
- [ ] I can distinguish what goes to the model, who saves the note and what I must decide.
- [ ] I can name a capability my next task needs and one I do not need yet.

You can answer on paper or in a text editor. No model account, connector or installation is needed.

## What each part does [#what-each-part-does]

| Component | What it does | Why it matters | Guide |
|---|---|---|---|
| Researcher | Brings the question, makes the judgments, checks the output | The only part that knows what the review is for | [Session 1](/docs/case-studies/systematic-reviews/session-1-review-by-hand) |
| Cherry Studio | Holds the conversation and the agents, and calls the tools and models below it | One window to compare models before an agent touches your files | [Cherry Studio](/docs/implementation/agentic-ai-tools/cherry-studio-setup-guide) |
| Zotero MCP | Lets the model search and read your Zotero library | Citations come from your library, not from the model's memory | [Zotero MCP](/docs/toolkit/zotero-mcp) |
| Filesystem MCP | Lets the model read and write files in the folders you allow | Scope it to the vault folder only, so the rest of your disk stays out of reach | [MCP Explorer](/docs/implementation/ai-environment-setup/mcp-explorer-guide#part-1-built-in-servers-for-cherry-studio) |
| Research Memex MCP server | Serves this site's pages and its three roles to the model, at `research-memex.org/mcp` | The model can look up a method or run a role without you pasting pages in; the roles run on your own model | [How AI reads this site](/docs/introduction/llms) |
| OpenRouter, Haimaker | Gateways: one account reaches models from several providers, so a course can buy credits once and issue each participant a capped key. The module has used OpenRouter and Haimaker; Vercel and Cloudflare run gateways of the same kind | Everyone can run the same task through different frontier models without opening accounts | [API Keys](/docs/implementation/ai-environment-setup/api-keys-setup-guide) |
| Local model | Runs a model on your own machine | Optional, free and private, and weaker than frontier models | [Free and low-cost options](/docs/implementation/core-references/ai-model-reference-guide#free-and-low-cost-options) |
| Zotero library | Holds your sources and their bibliographic records | The citation record that outlasts any conversation | [Zotero](/docs/implementation/foundational-setup/zotero-setup-guide) |
| Obsidian vault | Holds your notes, protocol and the AI's output as Markdown files | The place decisions are written down and can be found again | [Obsidian](/docs/implementation/foundational-setup/obsidian-setup-guide) |
| Papers as Markdown | Converted text of each PDF | A model reads and quotes Markdown more reliably than a PDF | [PDF to Markdown](/docs/implementation/ai-environment-setup/ocr-pdf-conversion-guide) |
| Research Rabbit | Finds related papers through citation networks | Widens the search from the papers you already trust | [Research Rabbit](/docs/implementation/foundational-setup/research-rabbit-setup-guide) |
| OpenAlex | An open index of the literature with an API and a CLI | A search an agent can run and you can rerun | [OpenAlex](/docs/implementation/foundational-setup/openalex-setup-guide) |

<span id="minimum-recommended-full" />

## Add tools for a task [#add-tools-for-a-task]

Add a capability when the task calls for it:

- read a supplied passage: a text editor is enough; an approved conversation interface can provide a second account
- search an existing library: add Zotero access when copying selected records becomes cumbersome and your host supports it
- read or save stored notes: add folder access only for the folders and operations you intend to allow
- handle a difficult PDF: convert it when needed, then check the converted text against the original
- retrieve this site's guidance: read its pages or connect the Research Memex MCP server; that server does not open your research files
- compare models or divide a task among agents: state what the comparison or division will help you decide before adding it

For a guided setup, [Carrel](/docs/toolkit/carrel) configures the machine with you; for the order to install things in, follow [Quick Start](/docs/implementation/foundational-setup/quick-start-checklist).

Erkko Autio's teaching feedback in October 2026 prompted this workflow-first explanation and its checks before setup.