Getting Oriented

Before you choose a tool or workflow, you need a map of the territory.

Welcome to the Research Memex, a guide to partnering with AI to amplify your thinking, not replace it. You're the driver. AI is the GPS.

This page is the map. It names what each of the five tabs holds and the page to open first in each, then sets out the ways in. The Contents lists every page in order. By Situation is the other map, arranged by what is in front of you rather than by tab.

How this site is organized

The site moves from the reasons for the approach, through setup and a worked case, to systems with more than one AI.

• Introduction: why this approach exists and the principles behind it, including interpretive orchestration, the mirror effect and research taste, with shorter essays on publishing and on choosing models. This page belongs here. Open Why Engage with AI in Research? first, then Core Principles.
• Implementation: the hands-on tab. It opens with The Research Workbench, a drawing of how the tools connect, and Quick Start, which orders the setup and helps you choose the smallest stack that serves you. Then come Model Access, the Core References on prompting, model choice and AI failure, three Roles for an AI you can hand to a model, the Essential Tools for references and notes, the AI Environment and the Agentic AI Tools. Open the Workbench first, then Quick Start.
• Case Studies: one worked case, a four-session course on systematic literature reviews. Running the Course holds the course overview, the student vault of worksheets and a practice set, and the instructor's decision on how the class reaches a model. The Sessions follow: Session 1 is the review done by hand; Sessions 2 to 4 bring AI into it.
• Toolkit: a catalog of agent plugins, skill artifacts, MCP servers and research platforms maintained alongside this site. Each page says what the tool does, which hosts it runs on and how to install it. Open the Toolkit index first: it says which tools recur in daily work and which are built for other people's use.
• Advanced Topics: two pieces on research systems with more than one AI. Read Composing an Agentic Research System first, for the questions that locate any arrangement, then Agentic Research Workflows for one working stack.

Choose a way in

Six starting orders. None is a curriculum: skip what you already know, and switch when another fits better. The first four are reading paths; the last two hand the first step to an AI.

Before installing tools on any reading path, follow one task through the Research Workbench and try its understanding check. You need no account or installation to see what moves between a source, a model and a checked note.

Learn the method through the case study

For researchers whose goal is a literature review or theory building, and for anyone taking the course. Instructors have their own path below.

1. Read the course overview, then work through Session 1, the review done by hand.
2. Read the three Core References: Cognitive Blueprint Prompts, the AI Model Reference Guide and the Failure Museum.
3. Follow your instructor's working mode, or choose one for independent practice. Use Quick Start if that mode needs tools.
4. For an AI run, choose an approved interface before configuring its model access and connections.
5. Keep your sources and notes in a place you can return to, then continue with Sessions 2 to 4. The Student Vault also supports peer work and individual reading without AI.

Teach the course

For instructors running the course or adapting it to their own field.

1. Read the course overview, including what the first two runs taught its authors, then work through Session 1 yourself.
2. Decide how the class reaches a model before Session 2: one gateway with a capped key per participant, cloud credits for a class that builds, or the access participants already hold. Test the route a week before the cohort starts.
3. Give participants the Student Vault and a course brief that names the working mode and the access route. The Quick Start then sends them to the right guide.
4. Run Sessions 2 to 4 at the pace your cohort needs. Point participants at the Failure Museum when an output looks wrong and they cannot yet say how.

Start at low cost

For keeping spending near zero while you find out what you need.

1. Try the walkthrough with a text editor or an interface you already have.
2. If you want a model response, consult the AI Model Reference Guide and choose a host through Quick Start.
3. Check the chosen route's access options before spending anything.
4. Add library, discovery or note tools only when the next task needs them.

Extend a working setup

For researchers who already work with AI daily and have a specific task their current setup cannot handle well.

1. Name the missing capability, its expected output and the check that would show it helped.
2. Use the Workbench's task-based additions to choose one change.
3. Test that change on a bounded task with non-sensitive inputs. Keep the earlier result for comparison.
4. Read Composing an Agentic Research System before adding another agent.

Be asked questions first

If you would rather describe your research than choose from a list, give the Research-AI Orientation role to an AI: paste the page's text into a conversation. It asks six to eight questions, one at a time, about how you work now, and ends with a short profile and one concrete possibility to try. Nothing is installed, and there is no score.

Have your machine configured

If you want the setup done with you instead of reading about it, Carrel interviews you, audits your machine and configures Obsidian, conversion tools and optional MCP servers. It runs as a Claude Code plugin or as a portable skill in other Agent Skills hosts, and installs a subset of the Toolkit. The Quick Start roadmap remains the guide to understanding what it set up.

Time investment

Knowing the time commitment helps you plan realistically.

• Setup: a one-time cost of two to three hours. Tool installation takes one to two hours, API configuration 30 to 60 minutes and first test runs 30 minutes. Follow the guides in order and do not skip their verification steps.
• The course: seven to nine hours a week while you follow the case study, split between 2-3 hours of preparation, 2 hours in session and 3-4 hours of practice afterwards.
• The return: expect an initial learning cost. Benefits vary with the task, your experience, and the time needed to check the output. Judge the setup by the work it helps you do, rather than expecting a fixed speedup or break-even week.

Our approach: one path among many

AI research methods include multiple valid approaches. Some focus on automation, with tools that handle specific tasks efficiently. Others focus on augmentation, with frameworks that amplify human thinking. Both have value for different contexts and goals.

We focus primarily on augmentation through what we call "interpretive orchestration." Our tool list is intentionally short: we treat the chosen tools as pedagogical instruments, not prescriptions, chosen to develop skills that transfer across contexts. We're still learning through our own research and teaching.

This approach may not work for you. That's okay.

A mindset for success

Treat the setup as an experiment: start small, expect mistakes, and use confusion as a signal that learning is in progress. A working setup is better than a perfect, overly complex one.

Contributing

The site's source repository is private for now: it is revised weekly with agents and carries working material that is not ready to publish. Contributions are welcome all the same, and the site is built to make them easy. Every page listed in /llms.txt has a Markdown twin (append .md to its URL) and /llms-full.txt holds the whole corpus, so any agent can draft a change from the real source. Send the files, or a folder of them, to the author (opens in a new tab); they will be merged with attribution. If contributions grow, a public repository follows.

---

The goal is to begin a journey of conscious competence in research thinking, not to master everything immediately.

@misc{docs-introduction-getting-oriented-2026,
  author = {Xule Lin},
  title = {Getting Oriented},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/introduction/getting-oriented}},
  note = {ORCID: 0000-0001-7885-4194}
}