The Simple Mode provides a streamlined AI coding workflow inside the normal left pane. It keeps the app structure familiar while reducing clutter.

You can activate it with the "Simple AI coding" switch just below the Sources bar. When you sign up and choose to have AI options switched on and active, Simple AI is turned on by default and you receive 10 free AI credits per month. AI usage consumes credits (see Responses Panel); credits renew monthly and do not roll over. Costs depend on model and workflow, but very roughly you might autocode around 30 pages for about 1 credit.

The Simple Workflow#

When Simple Mode is active:

• the Sources bar is hidden (to keep focus),
• the right-hand output tabs stay available,
• the Create/Filter sub-tabs are replaced by one combined simple workflow panel.

The workflow is broken down into six straightforward sections:

1. Run all: Optional one-click runner. Press Run all to run Auto-code, then Revise codebook, then Recode in order.
2. Runner pre-steps: clears Filter Links, turns filter pipeline on, and (if links already exist) asks once whether to delete all links before starting. You always start fresh.
3. Recode target suffix: Choose blank (simpler — synthesised labels go straight into cause/effect) or e.g. _recoded (keeps raw labels, writes synthesised to temp columns so you can compare).
4. Runner uses one top-level confirmation and suppresses the extra per-step confirm dialogs.
5. Background: Give the AI project context before coding. A status tick indicates whether enough background text is set.
6. Auto-code: This is where the AI reads your documents and extracts causal links.
7. You can choose to process a small sample first (e.g., 1 or 5 sources) to test your prompt, or process 100% of them.
8. The "Skip coded" switch ensures you don't waste time and money re-processing documents that already have links.
9. Default model in Simple AI is Qwen Flash.
10. Revise codebook: Once you have some causal links, the AI can review them and suggest a cleaner, more consistent list of factor labels (a "codebook"). The header tick shows whether the Recode codebook area currently contains suggestions.
11. Includes a Target clusters slider (2 to 50, default 20).
12. Optional Use automatic pre-clustering switch (default OFF).
13. When pre-clustering is OFF, the AI tries to find the clusters directly from the factor list using the standard Revise codebook prompt. This prompt supports macro replacement: use [number] (or [cluster_count]) and the slider value is injected at run time.
14. When pre-clustering is ON, the app first groups factor labels semantically using embeddings, then sends those clustered groups to the AI with a separate labelling prompt plus a Representatives per cluster slider (8 to 20, default 8).
15. Pre-clustering is more systematic than asking the AI to find all clusters "in its head" from a long raw list. It reduces the black-box / WEIRD-data risk a bit, and may make it easier to preserve more unusual or divergent concepts instead of collapsing them into whatever the model finds most typical.
16. Default model in Simple AI is Gemini 3 Flash Preview.
17. Recode: Apply the AI's suggested, cleaned-up labels back to your existing causal links. Paste the codebook (from Revise codebook or your own), add a recode instruction, and run.
18. The AI returns index mappings (row → codebook item) rather than full label text, reducing tokens and improving reliability.
19. Default instruction: "For each raw label give me the NUMBER of the best-matching codebook item by meaning. Use 0 when no codebook item fits. Return only codebook label numbers, never words. Never invent labels."
20. Skip recoded: When on, only processes links that have at least one unrecoded label (cause or effect). Use this when recoding again to focus on remaining work.
21. Links limit (1, 5, 20%, 50%, 100%): When not 100%, a random sample of links is recoded. Non-sampled links keep their existing recoded values (or stay blank on first run).
22. The header progress bar is segmented: grey = empty recoded fields, orange = recoded equals original cause/effect, green = recoded non-empty and different.
23. Default model in Simple AI is Qwen Flash.
24. Filter links: The normal Filter Links panel appears as the final section of the same accordion, so filtering is part of one continuous simple flow.
25. After a successful Run all, filters are auto-set to: Temporary Factor Labels (_recoded) → Factor Frequency (top 12) → Link Frequency (top 30).

Run all (Simple AI)#

• Optional sequencer for the three main AI actions.
• When enabled, Go runs Auto-code → Revise codebook → Recode, stopping on the first non-successful stage.
• Before running, it clears filters, enables pipeline, and deletes all existing links (if any) after one confirmation — so you always start from a clean slate.
• Recode target suffix (saved per project): After raw coding and Revise codebook, Recode applies synthesised factor labels. You can either:
• Blank: Write the synthesised labels straight into cause/effect. Simpler, but you lose the raw labels.
• _recoded (or another suffix): Keep the raw labels in cause/effect; write synthesised labels to temp columns (e.g. cause_recoded, effect_recoded). Lets you compare raw vs synthesised and switch between them via the Temporary Factor Labels filter.

Background (Simple AI)#

• Sets shared project context used by AI coding prompts.
• The status tick indicates whether enough background text is present.

Auto-code (Simple AI)#

• Runs AI coding across selected/all sources using your prompt and model.
• Use source limit + skip coded options to test quickly and avoid rework.
• Default model is Qwen Flash.

Revise codebook (Simple AI)#

• Suggests a cleaner consolidated codebook from existing links.
• Use this after you have enough coded links for a representative sample.
• Header tick indicates whether the Recode codebook area currently has content.
• Includes Target clusters slider (2-50, default 20).
• Optional Use automatic pre-clustering switch (default OFF).
• With pre-clustering OFF, the AI clusters the factor list directly from the Revise codebook prompt. That prompt supports [number] / [cluster_count].
• With pre-clustering ON, embeddings are used first to group labels semantically, then the AI only has to label those grouped clusters. This is a bit more systematic, less dependent on the AI doing all clustering internally as a black box, and may help preserve unusual or divergent concepts.
• Pre-clustering also adds a Representatives per cluster slider (8-20, default 8) and uses a separate labelling prompt.
• Default model is Gemini 3 Flash Preview.

Recode (Simple AI)#

• Applies your codebook back onto existing links, turning raw factor labels into cleaner synthesised ones.
• Recode target (set in Run all card): Blank = write straight to cause/effect (simpler). A suffix like _recoded = keep raw labels, write synthesised to temp columns (more flexible — compare raw vs synthesised).
• Supports sampled recoding and skip-recoded behavior (skip-recoded only applies when using a suffix).
• Header bar shows recode coverage mix across all cause/effect recoded fields.
• Default model is Qwen Flash.

Filter links (Simple AI)#

• This is the same Filter Links workflow, embedded as the final simple-ai accordion section.
• Use it to refine/select links before reviewing outputs on the right.
• Run-all completion auto-applies _recoded temp labels, top-12 factor frequency, then top-30 link frequency.

Advanced Settings#

Each section header is clickable and opens/collapses its settings panel. Section headers also include contextual Help buttons. The advanced sections are inline (not flyouts), and only one section is expanded at a time.

Inside advanced panels you can:

• Edit the exact Prompt the AI uses.
• View your prompt history and load previous prompts.
• Change the AI Model (e.g., switch to a "Pro" model for complex reasoning, or a "Flash" model for speed).
• Tweak technical settings like chunk size, concurrency, and temperature.