docs(TSP-1344): update model thinking/reasoning settings documentation by claude[bot] · Pull Request #721 · RelevanceAI/relevance-docs

claude · 2026-07-15T03:03:28Z

Summary

Documents the model thinking/reasoning settings on the Agent Language Model Settings page.

The Reasoning / Thinking section now covers:

Thinking type and effort level — set Thinking Type to Adaptive and choose an Effort Level, with the supported Claude models and a screenshot
Path to reach the settings: Agent → Settings → Advanced settings → Language Model
Cost impact warning: thinking tokens count as output tokens

Test plan

Verify the page renders correctly in Mintlify (image, callout)
Check all headings are sentence case
Confirm the <Warning> callout renders as a single paragraph

Linear issue: https://linear.app/relevance/issue/TSP-1344/

🤖 Generated with Claude Code

Substantially expand the Reasoning/Thinking sections on the Agent Language Model Settings page and the LLM Tool Step page with: - Legacy toggle explanation (non-functional, should be ignored) - Per-provider model support table (Anthropic, Google, OpenAI) - Detailed settings breakdown per provider including Claude Opus 4.5 Effort and GPT-5 Verbosity parameters - Known UI bug workaround for disappearing settings after model switch - Cost impact warning (thinking tokens = output tokens) - Adaptive thinking note - Feature history accordion Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

linear · 2026-07-15T03:03:31Z

TSP-1344

github-actions · 2026-07-15T03:07:32Z

🎯 Vibe check

Reviewed: 2 files (2 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Three heading-case errors in `llm-tool-step.mdx` (Advanced Settings, Fallback Model, Thinking / Reasoning Configuration); "LLM Tool Step" vs "Tool step" in the same sentence; "Fallback Model" capitalised in body text; model naming switches conventions mid-table in both files.
🟡	Technical clarity	7/10	Typo "choses" → "chosen" (`llm-tool-step.mdx:127`); two grammar errors; Fallback Model section in `llm-tool-step.mdx` omits the one-retry behaviour described in `language-model.mdx`.
🟡	Non-technical clarity	7/10	Opening sentence in `llm-tool-step.mdx` uses marketing framing ("inject the power of…"); "This is a good setting to select to make sure your Tool works!" is both vague and inappropriately casual.
🟢	Structure	9/10	Both pages are structured correctly for settings-reference use. Minor: numbered setup steps in `llm-tool-step.mdx` would benefit from `<Steps>`.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The new reasoning/thinking configuration sections are the best part of this PR — well-structured, consistent across both files, and the callouts, tables, and history accordions are all used correctly. The drag is in the pre-existing sections of llm-tool-step.mdx, which picked up several heading-case errors, grammar issues, and a typo when the new content was integrated. A quick cleanup pass on that file before merging would lift the overall quality noticeably.

🔧 Issues (11)

build/tools/tool-steps/llms/llm-tool-step.mdx:6 — "allows you to inject the power of large language models into your tools" is marketing framing. Rephrase to describe what the step does: "The LLM Tool step lets you use a language model as a processing step inside a Tool — for tasks like classification, summarization, and text generation."
build/tools/tool-steps/llms/llm-tool-step.mdx:7 — "The LLM Tool Step is a versatile Tool step" — "Step" should be lowercase "step"; inconsistent with the second usage in the same sentence.
build/tools/tool-steps/llms/llm-tool-step.mdx:41 — heading ## Advanced Settings → sentence case: ## Advanced settings
build/tools/tool-steps/llms/llm-tool-step.mdx:44 — heading ### Fallback Model → sentence case: ### Fallback model
build/tools/tool-steps/llms/llm-tool-step.mdx:45 — "This is a good setting to select to make sure your Tool works!" — exclamation mark is out of place in technical docs; "good setting to select" is also too vague. Describe what actually happens when the primary model fails.
build/tools/tool-steps/llms/llm-tool-step.mdx:46 — "Fallback Model" capitalised in body text should be "fallback model" — not a Relevance AI product proper noun.
build/tools/tool-steps/llms/llm-tool-step.mdx:50 — grammar: "notes, instruction and guides" → "notes, instructions, and guides"
build/tools/tool-steps/llms/llm-tool-step.mdx:68 — grammar: "help AI to know more about the situation" → "helps the AI model understand the conversation context" (or similar)
build/tools/tool-steps/llms/llm-tool-step.mdx:70 — heading ### Thinking / Reasoning Configuration → sentence case: ### Thinking / reasoning configuration
build/tools/tool-steps/llms/llm-tool-step.mdx:127 — typo: "the choses model" → "the chosen model"
build/agents/build-your-agent/agent-settings/language-model.mdx:69 — model naming is inconsistent within the same table row: "Claude 3.7 Sonnet", "Claude 4.0 Sonnet", "Claude 4.0 Opus" use version-before-name, but "Claude Opus 4.5" uses name-before-version. Anthropic's current convention is name-before-version (e.g. "Claude Sonnet 4.0", "Claude Opus 4.0"). The same table appears identically in llm-tool-step.mdx:84, so fix both.

🧩 Component suggestions (2)

build/tools/tool-steps/llms/llm-tool-step.mdx:32–38 — the five setup steps are a sequential procedure and would benefit from <Steps> instead of a plain numbered list. <Steps> adds visual progress indicators that help readers track where they are. Each step is already one action, so the migration is direct.
build/tools/tool-steps/llms/llm-tool-step.mdx:34–38 — UI element names are wrapped in single quotes ('LLM', 'Expand'). The convention used throughout these docs is **bold** for UI elements (e.g. **LLM**, **Expand**, **Prompt**). Align with that.

🏗️ Page structure (1)

build/tools/tool-steps/llms/llm-tool-step.mdx:44–45 — the Fallback Model section is a single vague sentence. language-model.mdx covers the same feature in much more depth: it explains the one-time retry behaviour, when to use a fallback, and the provider-diversity tip. Even a short version of those details here would make the Tool step doc self-contained. At minimum, mention that the fallback triggers automatically and only retries once.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~430

Files read: build/agents/build-your-agent/agent-settings/language-model.mdx (152 lines), build/tools/tool-steps/llms/llm-tool-step.mdx (153 lines), build/workforces/build-an-ai-workforce/edge-settings.mdx (120 lines), build/tools/tool-steps/llms/models.mdx (54 lines)

mintlify · 2026-07-15T03:09:39Z

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
relevanceai	🟢 Ready	View Preview	Jul 15, 2026, 3:09 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

mintlify · 2026-07-15T03:12:49Z

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
relevanceai	🟡 Building	–	Jul 15, 2026, 3:07 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

Verified against relevance-api-node model registry and builder-app gating: - Replace stale/incomplete reasoning support tables (were missing Claude Opus 4.6/4.7/4.8, Sonnet 4.6, Haiku 4.5, Gemini 3.x, all GPT-5.x variants) with capability-based framing that won't drift as models ship - Fix wrong Note claiming standard Claude Sonnet doesn't support reasoning (current Sonnets do — Sonnet 4.6 uses claude_thinking, 4/4.5 use token budget); the "disappears on Sonnet" behaviour is the UI bug, not lack of support - Remove false "GPT-5 verbosity may be feature-flagged for some accounts" claim — verbosity is gated by model capability (supports_verbosity) and the default agent skin, not any account/PostHog flag - Soften "Medium (default)" reasoning effort — default varies by model (GPT-5.1/5.2/5.4 default to none, GPT-5 Pro to high) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

github-actions · 2026-07-15T04:03:35Z

🎯 Vibe check

Reviewed: 2 files (2 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🔴	Consistency	5/10	`llm-tool-step.mdx` has three heading capitalization violations (`## Advanced Settings`, `### Fallback Model`, `### Thinking / Reasoning Configuration`), "ai" instead of "AI", verb form mismatch in steps, two grammar errors in body text, and inconsistent casing for "LLM Tool Step" vs the file's own title. `language-model.mdx` has one heading case slip (`### Reasoning / Thinking`).
🟡	Technical clarity	6/10	Both files share a confusing provider/model list that presents Claude Opus 4.5 and GPT-5 as peers of OpenAI, Google, and Anthropic rather than as specific models under those providers. `llm-tool-step.mdx` has two grammar errors that obscure meaning, a vague temperature section with no concrete examples, and "such as OpenAI" where a model name is needed. The fallback model retry limit (stated as "one time" in `language-model.mdx`) is absent from the equivalent section in `llm-tool-step.mdx`.
🟡	Non-technical clarity	7/10	`language-model.mdx` handles this well — "What it is" framing and a concrete temperature example. `llm-tool-step.mdx` drops context for temperature ("lose the right context" is unexplained), uses jargon ("hyperparameter") without definition, and says "know more about the situation" where a real explanation would help. The intro phrase "inject the power of" reads as filler rather than a definition.
🟡	Structure	7/10	The numbered steps in `llm-tool-step.mdx` (lines 33–38) should use `<Steps>` — sequential procedures get progress indicators that plain numbered lists don't. Everything else is structurally sound; both are settings/reference pages so no closing CTA is needed.

✨ Overall vibe: language-model.mdx is in solid shape — clear definitions, well-chosen callouts, and the reasoning section is thorough and consistent. llm-tool-step.mdx needs a pass before publish: it has multiple heading capitalization violations, grammar errors that undermine credibility, and several sections that read as first drafts (temperature, conversation history, fallback model). The reasoning/thinking sections that appear in both files are well-aligned and the most polished part of the PR.

🔧 Issues (21)

build/agents/build-your-agent/agent-settings/language-model.mdx

language-model.mdx:53 — ### Reasoning / Thinking → ### Reasoning / thinking (sentence case; "Thinking" is not a proper noun in this context)
language-model.mdx:83–87 — The provider list mixes providers (OpenAI, Google, Anthropic) with specific models (Claude Opus 4.5, GPT-5). Claude Opus 4.5 is an Anthropic model and GPT-5 is an OpenAI model. The flat list reads as five parallel providers. Add a note or restructure as sub-bullets to clarify: "Some models have additional parameters: Claude Opus 4.5 uses Effort; GPT-5 uses Verbosity."

build/tools/tool-steps/llms/llm-tool-step.mdx

llm-tool-step.mdx:4 — description: "Use a LLM as a step in your Tools" → "Use an LLM as a step in your Tools" ("LLM" starts with a vowel sound)
llm-tool-step.mdx:6 — "LLM Tool Step" → "LLM Tool step" (inconsistent with the page title, which uses lowercase "step")
llm-tool-step.mdx:6 — "inject the power of large language models" — "power" leans marketing-y; consider "lets you use large language models directly in your Tools"
llm-tool-step.mdx:33 — Step 1 uses a gerund ("Creating a new Tool") while steps 2–5 use imperatives ("Click", "Add", "Choose"). Align to imperative: "Create a new Tool, then click 'LLM' under Steps"
llm-tool-step.mdx:40 — ## Advanced Settings → ## Advanced settings (sentence case; "settings" is not a proper noun)
llm-tool-step.mdx:44 — ### Fallback Model → ### Fallback model (sentence case)
llm-tool-step.mdx:46 — "Fallback Model" (body text) → "fallback model"
llm-tool-step.mdx:46 — "This is a good setting to select to make sure your Tool works!" — remove the exclamation mark; rephrase to say what it actually does: "Use this to reduce failures caused by model availability issues."
llm-tool-step.mdx:50 — Grammar: "composed of notes, instruction and guides" → "notes, instructions, and guides". The trailing clause "or to follow a specific format, or limitations" doesn't parse; rewrite the sentence.
llm-tool-step.mdx:52 — "System prompt" is capitalized here but the section heading on line 49 is ### System prompt (sentence case). Drop the capital: "An example of a system prompt would be:"
llm-tool-step.mdx:56 — "LLMs' response" → "LLM's response" (the LLM is singular in this context)
llm-tool-step.mdx:60 — "responses might also lose the right context" — vague; clarify what this means: e.g., "responses may drift off-topic or ignore constraints in your prompt"
llm-tool-step.mdx:64 — "Click on + Add row" → "Click + Add row" (CLAUDE.md: drop "on"); also "ai" → "AI" — AI is an acronym
llm-tool-step.mdx:66 — <Info>The only acceptable roles are "ai" and "user".</Info> — if "ai" is the literal API/UI value, keep it; if it's just a description, capitalize to "AI"
llm-tool-step.mdx:68 — "help AI to know more about the situation" → "helps the model understand the prior conversation context" ("help" subject-verb disagreement; "know more about the situation" is vague)
llm-tool-step.mdx:70 — ### Thinking / Reasoning Configuration → ### Thinking / reasoning configuration (sentence case; neither "Reasoning" nor "Configuration" is a proper noun here)
llm-tool-step.mdx:124 — "For certain models such as OpenAI" — OpenAI is a provider, not a model. Rewrite: "For certain models such as GPT-4o" or "For models from certain providers such as OpenAI"
llm-tool-step.mdx:129 — "The error message below indicate that the provided prompt includes more tokens than what the choses model allows." → "indicates" (subject-verb agreement); "choses" → "chosen"
llm-tool-step.mdx:135 — Accordion title "Too large data" is grammatically wrong → "Input data too large". Also the body text has "you need to upload it as a dataset use it as a knowledge" — missing "and": "upload it as a dataset and use it as Knowledge in your Tool"

🧩 Component suggestions (1)

llm-tool-step.mdx:33–38 — The 5-item numbered list for adding an LLM Tool step is a sequential procedure; use <Steps> instead of a plain numbered list. <Steps> adds visual progress indicators and is the Mintlify convention for step-by-step instructions. The resulting block would wrap each numbered item in a <Step title="..."> tag with an imperative title (e.g., <Step title="Create a new Tool">).

⚠️ Contradictions (1)

language-model.mdx:137 states the fallback model retries failed tasks "one time" and that if the fallback also fails, the task fails as usual. llm-tool-step.mdx:44–46 describes the Fallback Model setting without mentioning the one-retry limit. If the behavior is the same (which it likely is), the tool step page should say so — otherwise readers of the tool step page will not know the retry is capped at one attempt.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	3
Total lines processed	~732

Files read: language-model.mdx (154 lines), llm-tool-step.mdx (155 lines), general.mdx (249 lines), edge-settings.mdx (120 lines), models.mdx (54 lines)

…llout Scope the additions to what the source Slack thread actually covered: legacy toggle is non-functional, reasoning settings are per-model, thinking counts as output tokens (cost), and Claude adaptive thinking / effort levels. - Remove the "Known UI issue" callout (settings disappearing on model switch) from both pages — a live bug shouldn't be enshrined as docs - Remove the "Reasoning feature history" accordions (changelog content, unverified dates — belongs in the changelog, not a settings page) - Fold the model support table and standalone "adaptive thinking" heading into the per-model prose to cut length Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

github-actions · 2026-07-15T05:24:33Z

🎯 Vibe check

Reviewed: 2 files (2 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Three sentence-case heading violations in `llm-tool-step.mdx`; one in `language-model.mdx`; single quotes used instead of bold for UI element names; "instruction" (singular) and "choses" typos; "knowledge" (product feature) lowercased.
🟡	Technical clarity	6/10	Provider/model list in `language-model.mdx` conflates specific model names with provider dropdown options; two grammar errors in `llm-tool-step.mdx` ("indicate" / "choses model") make sentences misleading; missing "and" in line 110 creates a run-on that loses meaning; no mention of the Provider dropdown in the tool-step version of the reasoning config.
🟡	Non-technical clarity	7/10	`llm-tool-step.mdx` temperature section uses "hyperparameter" with no definition, then ends on "responses might also lose the right context" — vague enough to confuse. The casual "This is a good setting to make sure your Tool works!" undercuts trust. `language-model.mdx`'s temperature treatment is noticeably cleaner.
🟡	Structure	7/10	Procedural steps in `llm-tool-step.mdx` are a plain numbered list where `<Steps>` is the right component; the Fallback Model section there is one thin sentence with no "when to use" context, whereas `language-model.mdx` covers retry behavior, provider diversity, and edge cases.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The new thinking/reasoning sections are the strongest part of this PR — the content is accurate, the warning callouts are correctly typed, and the two pages stay consistent with each other on the key facts. The drag is that llm-tool-step.mdx carries pre-existing quality debt (grammar errors, jargon, casual tone, capitalization violations) that this PR touched but didn't resolve, and the conflation of specific model names with provider options in language-model.mdx needs untangling.

🔧 Issues (12)

build/agents/build-your-agent/agent-settings/language-model.mdx:53 — heading ### Reasoning / Thinking violates sentence case — should be ### Reasoning / thinking
build/agents/build-your-agent/agent-settings/language-model.mdx:65–69 — "Claude Opus 4.5" and "GPT-5" are listed as bullet items in the same format as provider names (OpenAI, Google, Anthropic), but they're specific models. The intro sentence says "Using the Provider dropdown, select the provider that matches your model" — listing model names as if they're providers is confusing. Either label them clearly as model-specific overrides ("For Claude Opus 4.5 specifically:…") or separate them from the provider list.
build/tools/tool-steps/llms/llm-tool-step.mdx:40 — ## Advanced Settings → ## Advanced settings (sentence case; "Settings" is not a proper noun here)
build/tools/tool-steps/llms/llm-tool-step.mdx:44 — ### Fallback Model → ### Fallback model (sentence case)
build/tools/tool-steps/llms/llm-tool-step.mdx:70 — ### Thinking / Reasoning Configuration → ### Thinking / reasoning configuration (sentence case)
build/tools/tool-steps/llms/llm-tool-step.mdx:34–35 — UI element names use single quotes instead of bold: clicking 'LLM' under Steps → clicking **LLM** under Steps; Click 'Expand' → Click **Expand**
build/tools/tool-steps/llms/llm-tool-step.mdx:46 — "This is a good setting to select to make sure your Tool works!" — exclamation mark and casual framing are out of place in technical docs. Rewrite as: "Selecting a fallback reduces task failures caused by temporary provider outages."
build/tools/tool-steps/llms/llm-tool-step.mdx:50 — Grammar: "notes, instruction and guides" → "notes, instructions, and guides"
build/tools/tool-steps/llms/llm-tool-step.mdx:56–60 — "Temperature is a hyperparameter" drops jargon with no definition; "responses might also lose the right context" is too vague to be actionable. Compare language-model.mdx's range-based breakdown (0–0.3 / 0.4–0.7 / 0.8–1.0) which actually tells readers what to do.
build/tools/tool-steps/llms/llm-tool-step.mdx:68 — Grammar: "and help AI to know more about the situation" → "and helps the model understand the conversation context"
build/tools/tool-steps/llms/llm-tool-step.mdx:103 — Grammar: "The error message below indicate that" → "indicates that"; "the choses model" → "the chosen model"
build/tools/tool-steps/llms/llm-tool-step.mdx:110 — Missing "and": "upload it as a dataset use it as a [knowledge]" → "upload it as a dataset and use it as [Knowledge]"; also lowercase "knowledge" should be "Knowledge" (Relevance AI product feature)

🧩 Component suggestions (1)

build/tools/tool-steps/llms/llm-tool-step.mdx:32–36 — The numbered list ("1. Creating a new Tool… 2. Click 'Expand'…") is a sequential procedure and should use <Steps>. It also has inconsistent verb forms (gerund on step 1, imperative on steps 2–5) that <Steps> would naturally prompt the author to fix.

🏗️ Page structure (1)

build/tools/tool-steps/llms/llm-tool-step.mdx:44–47 — The Fallback Model section is a single vague sentence plus a filler line. The same setting in language-model.mdx covers: when to use it, that the fallback retries exactly once, what happens if the fallback also fails, and a tip about choosing a different provider. Either expand with the same detail, or add a cross-link: "See Language Model settings for guidance on when to use a fallback and how to choose one."

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	3
Total lines processed	~470

Files read: language-model.mdx (124 lines), llm-tool-step.mdx (129 lines), general.mdx (249 lines), edge-settings.mdx (120 lines), models.mdx (54 lines)

- Add the Advanced Settings > Language Model "Thinking" screenshot (images/agent/advanced-settings-thinking.png) - New "Thinking type and effort level" subsection on the agent Language Model page using the real UI labels (Thinking Type: Adaptive, Effort Level: Low/Medium/High up to Max), listing the Claude models that show these controls (Opus 4.6/4.7/4.8, Sonnet 4.6) - Link the LLM Tool Step page to that section instead of duplicating - Drop the GPT-5 verbosity bullet from both pages (out of scope) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

github-actions · 2026-07-15T06:00:44Z

🎯 Vibe check

Reviewed: 2 files (2 with issues, 0 clean) + 1 image asset (no review needed)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Heading case errors across both files (`## Advanced Settings`, `### Fallback Model`, `### Thinking / Reasoning Configuration` in llm-tool-step.mdx; `### Reasoning / Thinking` in language-model.mdx). Grammar errors throughout llm-tool-step.mdx: "a LLM", "instruction", "help AI", "indicate", "choses model".
🟡	Technical clarity	7/10	Claude Opus 4.5 listed as a separate bullet from the Anthropic provider without explaining why it diverges from the Thinking budget approach; "certain models such as OpenAI" (line 97) conflates providers with models; new screenshot missing `alt` attribute.
🟡	Non-technical clarity	7/10	The `What it is: / When to use it: / How to configure:` pattern in language-model.mdx is excellent. llm-tool-step.mdx is weaker — inconsistent UI element formatting (single quotes vs. bold), grammar errors that undercut credibility, and a vague Force response format section.
🟡	Structure	7/10	Sequential numbered steps in llm-tool-step.mdx should use `<Steps>`. Otherwise both pages follow appropriate reference page patterns.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The new Thinking/Reasoning configuration sections are the PR's core contribution and they're solid — well-structured, consistent between the two files, and the cross-link from llm-tool-step.mdx to the agent settings anchor resolves correctly. The main drag is a cluster of pre-existing grammar errors and heading case violations in llm-tool-step.mdx that the PR touches but didn't fix; since the PR edits this file, it's worth cleaning them up while you're in there.

🔧 Issues (11)

build/tools/tool-steps/llms/llm-tool-step.mdx:3 — 'Use a LLM as a step' → "an LLM" ("LLM" starts with a vowel sound)
build/tools/tool-steps/llms/llm-tool-step.mdx:41 — ## Advanced Settings → ## Advanced settings (sentence case)
build/tools/tool-steps/llms/llm-tool-step.mdx:44 — ### Fallback Model → ### Fallback model (sentence case)
build/tools/tool-steps/llms/llm-tool-step.mdx:50 — "instruction" → "instructions"; "guiding AI to assume a certain role, or to follow a specific format, or limitations" — parallel structure broken, or limitations doesn't match the to [verb] pattern; rewrite as "to follow specific limitations"
build/tools/tool-steps/llms/llm-tool-step.mdx:68 — "help AI to know" → "helps AI know"
build/tools/tool-steps/llms/llm-tool-step.mdx:70 — ### Thinking / Reasoning Configuration → ### Thinking / reasoning configuration (sentence case)
build/tools/tool-steps/llms/llm-tool-step.mdx:97 — "For certain models such as OpenAI" — OpenAI is a provider, not a model. Rewrite: "For certain models, such as those from OpenAI"
build/tools/tool-steps/llms/llm-tool-step.mdx:102 — "indicate" → "indicates"; "choses model" → "chosen model"
build/tools/tool-steps/llms/llm-tool-step.mdx:109 — "upload it as a dataset use it" → "upload it as a dataset and use it" (missing "and")
build/agents/build-your-agent/agent-settings/language-model.mdx:53 — ### Reasoning / Thinking → ### Reasoning / thinking (sentence case: "Thinking" after the slash is not a new sentence and isn't a product proper noun)
build/agents/build-your-agent/agent-settings/language-model.mdx:65–67 — <img> tag is missing an alt attribute; add descriptive alt text, e.g. alt="Thinking Type set to Adaptive with Effort Level dropdown showing Low, Medium, High options"

🧩 Component suggestions (1)

build/tools/tool-steps/llms/llm-tool-step.mdx:32–38 — The numbered list "1. Creating a new Tool … 5. Click the Settings icon" is a sequential procedure; CLAUDE.md says use <Steps> for these. Replace with a <Steps> block so readers get visual progress indicators. Each step should also use bold for UI element names instead of single quotes — e.g. Click **LLM** under Steps not clicking 'LLM' under Steps.

⚠️ Contradictions (1)

build/agents/build-your-agent/agent-settings/language-model.mdx:70–83 — The "Thinking type and effort level" section (lines 70–72) states the Adaptive/Effort Level controls appear on Claude Opus 4.6/4.7/4.8 and Claude Sonnet 4.6. Then the "How to configure" list (line 80) says Anthropic: enter a Thinking budget ≥ 1024, and immediately below (line 81) lists Claude Opus 4.5 as a separate bullet with a different control (the Effort parameter). It's unclear whether Opus 4.5 is accessed through the Anthropic provider dropdown or elsewhere, and whether it shares the Thinking budget mechanism or bypasses it. The same ambiguity appears in build/tools/tool-steps/llms/llm-tool-step.mdx:84–85. Both files are internally consistent with each other, but a reader with Opus 4.5 will be confused about which configuration path to follow. A clarifying parenthetical like "(Claude Opus 4.5 uses a separate Effort parameter — it does not use the Thinking budget)" would resolve this.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	3
Total lines processed	~490

Files read: build/agents/build-your-agent/agent-settings/language-model.mdx (136 lines), build/tools/tool-steps/llms/llm-tool-step.mdx (128 lines), build/workforces/build-an-ai-workforce/edge-settings.mdx (120 lines), build/agents/build-your-agent/agent-settings/general.mdx (249 lines), build/tools/tool-steps/llms/models.mdx (54 lines)

…add settings path Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

github-actions · 2026-07-17T01:42:51Z

🎯 Vibe check

Reviewed: 2 files (2 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Three heading capitalization errors in `llm-tool-step.mdx` (Advanced Settings, Fallback Model, Thinking / Reasoning Configuration); "LLM Tool Step" in body contradicts frontmatter "LLM Tool step"; grammatical error "a LLM" in description frontmatter.
🟡	Technical clarity	7/10	Missing `alt` attribute on the new screenshot in `language-model.mdx`; three UI element names formatted with single quotes or backticks instead of bold; Anthropic/Claude Opus 4.5 reasoning bullets poorly differentiated.
🟢	Non-technical clarity	9/10	Content is accessible. "inject the power of" in the llm-tool-step.mdx lede leans marketing-y but no banned words used.
🟢	Structure	9/10	Both pages fit the reference/settings pattern correctly. Numbered steps in `llm-tool-step.mdx` could use `<Steps>` for visual progress indicators, but it's minor polish.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The language-model.mdx changes are clean and well-structured — the new Thinking section is clear and appropriately scoped. The llm-tool-step.mdx changes introduce good content but bring several pre-existing hygiene issues into scope: three heading capitalization violations, inconsistent UI element formatting, and a confusing Anthropic model bullet that blurs the boundary between the "Thinking budget" setting and the Claude Opus 4.5 "Effort" parameter.

🔧 Issues (10)

build/tools/tool-steps/llms/llm-tool-step.mdx:2 — Description frontmatter: "Use a LLM as a step" → "Use an LLM as a step" (article before acronym starting with a vowel sound)
build/tools/tool-steps/llms/llm-tool-step.mdx:6 — Body text says "The LLM Tool Step" (capital S), but the frontmatter title is 'LLM Tool step' (lowercase s). Pick one and use it consistently throughout.
build/tools/tool-steps/llms/llm-tool-step.mdx:40 — ## Advanced Settings → ## Advanced settings (sentence case; "Settings" is not a proper noun)
build/tools/tool-steps/llms/llm-tool-step.mdx:44 — ### Fallback Model → ### Fallback model (sentence case; see sibling language-model.mdx:102 which already uses ### Fallback model)
build/tools/tool-steps/llms/llm-tool-step.mdx:70 — ### Thinking / Reasoning Configuration → ### Thinking / Reasoning configuration (sentence case)
build/tools/tool-steps/llms/llm-tool-step.mdx:34 — clicking 'LLM' under Steps → clicking **LLM** under **Steps** (UI element names should be bold, not single-quoted)
build/tools/tool-steps/llms/llm-tool-step.mdx:35 — Click 'Expand' → Click **Expand**
build/tools/tool-steps/llms/llm-tool-step.mdx:64 — Click on `+ Add row` → Click **+ Add row** (backtick wrapping a UI element name; use bold instead)
build/agents/build-your-agent/agent-settings/language-model.mdx:61–63 — <img src="/images/agent/advanced-settings-thinking.png" /> is missing an alt attribute. Add something like alt="Agent advanced settings showing Thinking Type and Effort Level controls".
build/tools/tool-steps/llms/llm-tool-step.mdx:84–85 — The Anthropic bullet says "Enter a Thinking budget — must be 1024 or above", then a separate bullet says "Claude Opus 4.5: Configure the Effort parameter." Claude Opus 4.5 is an Anthropic model, so the two bullets appear to conflict. Clarify the relationship: e.g., nest the Claude Opus 4.5 note under the Anthropic bullet as: "For Claude Opus 4.5, configure the Effort parameter instead of a Thinking budget."

🧩 Component suggestions (1)

build/tools/tool-steps/llms/llm-tool-step.mdx:33–38 — The five-item numbered list under "Add the LLM Tool step to your Tool" is a sequential procedure. Replace the markdown ordered list with a <Steps> component to get the visual progress indicators that distinguish procedures from lists. Example:

<Steps>
  <Step title="Create a new Tool">Click **LLM** under **Steps**.</Step>
  <Step title="Expand the step">Click **Expand** to see the full Tool step.</Step>
  <Step title="Add a prompt">Enter your prompt in the **Prompt** field.</Step>
  <Step title="Choose a model">Select the model from the model dropdown.</Step>
  <Step title="Access advanced settings (optional)">Click the **Settings** icon.</Step>
</Steps>

⚠️ Contradictions (1)

build/agents/build-your-agent/agent-settings/language-model.mdx:69 lists Thinking Type / Effort Level as available on Claude Opus 4.6, 4.7, 4.8 and Claude Sonnet 4.6. build/tools/tool-steps/llms/llm-tool-step.mdx:85 separately calls out Claude Opus 4.5 as supporting an "Effort" parameter. Claude Opus 4.5 appears nowhere in the agent-settings list. Readers comparing both pages will be unsure whether Opus 4.5 supports reasoning in Tool steps but not Agent settings, or whether the agent page simply omitted it. Align the model lists or add a note explaining the difference in support scope between the two surfaces.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	3
Total lines processed	~672

Files read: language-model.mdx (121 lines), llm-tool-step.mdx (128 lines), edge-settings.mdx (120 lines), general.mdx (249 lines), models.mdx (54 lines)

github-actions · 2026-07-17T01:43:18Z

🎯 Vibe check

Reviewed: 1 MDX file (1 with issues) + 1 image file (no text content to review)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	8/10	Two headings in title case instead of sentence case.
🟢	Technical clarity	9/10	Missing `alt` attribute on the new screenshot (`<img src="/images/agent/advanced-settings-thinking.png" />`).
🟢	Non-technical clarity	9/10	Writing is accessible and plain-English throughout. Minor: "Thinking type and effort level" section drops readers into control names without a one-sentence definition of what reasoning/thinking actually is.
🟢	Structure	9/10	First heading `### Language Model` repeats the page title — redundant against the sibling page pattern.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: A clean, well-structured settings reference page. The Thinking section is tightly written and appropriately scoped. Two heading capitalization fixes are the only must-fix items before merging.

🔧 Issues (3)

language-model.mdx:6 — ### Language Model → ### Language model. "Language Model" is a generic technical term, not a Relevance AI product name (which would be Agent, Tool, Workforce, etc.). Sentence case applies.
language-model.mdx:78 — ### Parallel Tool Calls → ### Parallel tool calls. "Tool Calls" here refers to LLM tool-calling behavior, not the Relevance AI "Tool" product feature. Sentence case applies.
language-model.mdx:61–63 — <img> tag has no alt attribute. Add a short description (e.g., alt="Thinking Type and Effort Level controls in the Language Model advanced settings panel") for accessibility and for when the image fails to load.

🏗️ Page structure (1)

language-model.mdx:6 — The first heading ### Language Model mirrors the page title exactly. The sibling page (general.mdx) goes directly into its first setting (### Suggest Replies) without a redundant top-level section header. Either rename this heading to describe what the section actually covers (### Model selection) or remove it and let the description metadata carry the intro weight.

✅ Clean files (2 context pages)

build/agents/build-your-agent/agent-settings/general.mdx (read for context — no issues found relevant to this PR), build/workforces/build-an-ai-workforce/edge-settings.mdx (read for contradiction check — no conflicts found)

🔋 Credit usage

Item	Count
Files reviewed	1 MDX + 1 image
Context pages read	2
Total lines processed	~490

Files read: language-model.mdx (121 lines), general.mdx (249 lines), edge-settings.mdx (120 lines)

claude Bot added the docs-drafter Documentation drafted by Claude label Jul 15, 2026

claude Bot assigned NiamhRelevance Jul 15, 2026

mintlify Bot deployed to staging July 15, 2026 03:09 View deployment

mintlify Bot deployed to staging July 15, 2026 04:02 View deployment

mintlify Bot deployed to staging July 15, 2026 05:24 View deployment

mintlify Bot deployed to staging July 15, 2026 05:59 View deployment

docs(language-model): trim thinking section to type/effort and cost, …

6aac825

…add settings path Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

NiamhRelevance requested a review from jordanc-relevanceai July 17, 2026 01:39

mintlify Bot deployed to staging July 17, 2026 01:40 View deployment

docs: drop LLM Tool step changes from PR, keep only language-model page

2cb636e

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

mintlify Bot deployed to staging July 17, 2026 01:41 View deployment

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

docs(TSP-1344): update model thinking/reasoning settings documentation#721

docs(TSP-1344): update model thinking/reasoning settings documentation#721
claude[bot] wants to merge 6 commits into
mainfrom
docs/TSP-1344

claude Bot commented Jul 15, 2026 •

edited by NiamhRelevance

Loading

Uh oh!

linear Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026

Uh oh!

mintlify Bot commented Jul 15, 2026

Uh oh!

mintlify Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 17, 2026

Uh oh!

github-actions Bot commented Jul 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

claude Bot commented Jul 15, 2026 • edited by NiamhRelevance Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

linear Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026

🎯 Vibe check

Uh oh!

mintlify Bot commented Jul 15, 2026

Uh oh!

mintlify Bot commented Jul 15, 2026

Uh oh!

github-actions Bot commented Jul 15, 2026

🎯 Vibe check

Uh oh!

github-actions Bot commented Jul 15, 2026

🎯 Vibe check

Uh oh!

github-actions Bot commented Jul 15, 2026

🎯 Vibe check

Uh oh!

github-actions Bot commented Jul 17, 2026

🎯 Vibe check

Uh oh!

github-actions Bot commented Jul 17, 2026

🎯 Vibe check

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

claude Bot commented Jul 15, 2026 •

edited by NiamhRelevance

Loading