docs(61cedc90-3290-470f-8c20-158b3ecd83ef): update Invent v3 features documentation

[claude]

Summary

• Removes the incorrect Enterprise-only access warning — Inventor access is role-gated (operator+), not plan-gated
• Documents six new Invent v3 features as tabs in the "Inventor features" section: Memories, Scheduled check-ins, Voice sessions, Subagent spawning, Context-aware suggested actions, and Tool approval queue
• Updates the Model selection tab to cover the three-tier model grouping (Cheapest / Balanced / Most capable) and per-plan defaults
• Adds two new cards to "What you can do": voice-based building and parallel task execution via subagent spawning
• Clarifies that the standalone /invent route is being deprecated and the universal side panel is the primary access method

Rollout status

All features documented here are live at 100% to pilot cohorts 27933 and "Invent + Evals Pilot (Mid 2026)". Features verified against apps/builder-app/features/invent/v3/.

Source

Requested by Jordan to align docs with new Invent v3 features in the repo.

Linear: https://linear.app/relevance/issue/61cedc90-3290-470f-8c20-158b3ecd83ef/

[mintlify]

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

Project	Status	Preview	Updated (UTC)
relevanceai	🟢 Ready	View Preview	Aug 13, 2026, 2:27 AM

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

[github-actions]

🎯 Vibe check

Reviewed: 1 file (1 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟢	Consistency	9/10	No banned words, no British spellings, sentence case holds throughout. Minor: "LLM" on line 174 is undefined, and "Invent session" (line 174) is inconsistent with "Inventor" used everywhere else.
🟡	Technical clarity	8/10	Good UI specificity overall. Callout severity mismatches: <Info> is used for a deprecation notice (line 56–58) and for permanent memory deletion (line 206–208) — both warrant <Warning>. Credit cost mentioned on line 138 but never quantified.
🟡	Non-technical clarity	7/10	"LLM" on line 174 is used without definition. "fan-out execution" on line 44 is developer jargon dropped without explanation. "MCP" appears on line 11 before it's defined (the dedicated section is at the bottom of the page).
🟡	Structure	7/10	No closing CTA on a concept/overview page. Three <CardGroup> blocks are used for descriptive properties rather than navigable options — the two- and three-item property lists would read better as prose.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5. Scores are a single overall judgment about the whole PR — not per file. The dropdowns below contain the line-by-line specifics.

✨ Overall vibe: This is a well-written, comprehensive Inventor reference page with accurate product-term capitalization, no banned words, and genuinely useful specifics like the credit counter, auto-accept behavior, and version restore gotcha. The main gaps are a missing closing CTA, a few callout types used at the wrong severity level, and some two-to-three item <CardGroup> blocks that are overkill for what are essentially short descriptive lists.

🔧 Issues (5)

• build/invent/invent.mdx:56–58 — <Info> used for a deprecation notice ("The standalone /invent route is being deprecated"). Deprecation is a warning-level concern — users acting on outdated info may hit broken routes. Change to <Warning>.
• build/invent/invent.mdx:174 — "The model dropdown sets which LLM powers your Invent session." Two issues: (1) "LLM" is undefined — non-technical readers won't know the acronym; spell out "large language model (LLM)" on first use. (2) "Invent session" should be "Inventor session" to stay consistent with every other reference on the page.
• build/invent/invent.mdx:206–208 — <Info> callout says "Deleting a memory removes it permanently." Permanent deletion is an irreversible action — per CLAUDE.md, this is exactly what <Warning> is for. Change to <Warning>.
• build/invent/invent.mdx:318–320 — <Info> callout says "Closing the panel or refreshing the page doesn't delete your conversation history." This is a positive reassurance, not neutral information. Change to <Check>.
• build/invent/invent.mdx:44 — "enabling fan-out execution for complex, multi-step workflows" in the "Run tasks in parallel" card description. "Fan-out execution" is developer jargon. The Tab section on line 245 explains it well in plain language — use similar phrasing here, e.g. "so complex tasks run faster by splitting across multiple workers."

🧩 Component suggestions (3)

• build/invent/invent.mdx:48–50 — <Warning> wraps "Inventor can log bugs on your behalf. These go to our engineering team, not support, so you won't receive updates on their progress." This is informational, not a risk or irreversible action — there's nothing dangerous about Inventor logging a bug. Change to <Note>.
• build/invent/invent.mdx:292–302 — <CardGroup cols={3}> holds three descriptive properties: "Tone and language", "Domain assumptions", "Personalization". These are characteristics of the About You feature, not navigable options or parallel choices the reader picks between. Per CLAUDE.md, short property lists work better as prose or bullets. Replace with a short bulleted list, e.g.: "• More precise technical language for engineers, simpler explanations for non-technical users • Domain-aware defaults (e.g. sales-flavored suggestions when your role is in sales) • First-name address at natural moments like greetings".
• build/invent/invent.mdx:325–331 — <CardGroup cols={2}> holds exactly two items: "Thumbs up" and "Thumbs down". Two-item card groups for binary actions add visual overhead without helping comprehension. Roll these into the paragraph above: "Thumbs up marks a response as accurate and helpful; thumbs down opens a bug report form where you can describe what you asked for, what Inventor generated, and what was wrong."

🏗️ Page structure (1)

• build/invent/invent.mdx — No closing CTA. This is a concept/overview page — per CLAUDE.md, it should end with a "What's next?" pointing readers somewhere actionable now that they know what Inventor is. Suggested links: /build/agents/build-your-agent/overview (build your first agent), /build/tools/create-a-tool (already referenced inline — make it a formal next step), and /integrations/mcp/mcp-server (already explained in the MCP vs Inventor section — round it out for readers who want to go further).

✅ Clean files (0)

No files were fully clean — build/invent/invent.mdx had issues across multiple dimensions.

🔋 Credit usage

Item	Count
Files reviewed	1
Context pages read	2
Total lines processed	~795

Files read: build/invent/invent.mdx (361 lines), build/tools/create-a-tool.mdx (102 lines), integrations/mcp/mcp-server.mdx (332 lines)

[NiamhRelevance]

Change Inventor to Invent

[NiamhRelevance]

The section called inventor features is too long now. Let's think how we can change this component and still make it look good

[NiamhRelevance]

Add a scheduled check in section in the Invent features section. Here is some info: Invent can check back in on your work later. Ask it to review an agent in a few days and it will schedule a message to itself, then look again at the time you agreed. If it finds something worth your attention it lets you know. If everything looks fine, it stays quiet. Schedule a check-in

Just ask. There's no form to fill in. Tell Invent to check back on an agent later and it will suggest a sensible interval based on how busy that agent is, from about a day for agents running many tasks a day, up to a week for quiet ones. You can pick a different time if you'd rather. What happens when a check-in runs

Invent reviews the resource the way you would: test results and monitoring first, then errors, then recent activity. If nothing needs attention, you won't hear from it. If something does, it sends you a short summary of the problem along with a suggested fix. Approve the fix and Invent applies it to your draft, then offers to check back again to confirm it worked.

[NiamhRelevance]

remove this: The standalone /invent route is being deprecated. Use the side panel instead.

[NiamhRelevance]

Add this if you haven't already: Invent will be available by default for all org admins, and no one else. For each project, they will have a toggle to turn it on for all users in that project

[NiamhRelevance]

https://relevanceai-docs-invent-v3-features.mintlify.site/build/invent/invent#what-you-can-do clean up this component. its quite long now. we also need to update it based off marketing info in here: https://relevance-site-git-invent-page.preview.relevanceai.com/invent

[NiamhRelevance]

Put this into a FAQ: Inventor can log bugs on your behalf. These go to our engineering team, not support, so you won’t receive updates on their progress.

[NiamhRelevance]

add this: when you click on the models, you can see the available models that can or cannot include text and images

[NiamhRelevance]

Alarms / alerts. will look at your evals and make suggestions. Set it up in the same way as scheudled checkins. Set the threshold, can configure sensitivity, and invent can set this all up for you - just ask it. Look up in the code how to access these steps

[NiamhRelevance]

include screenshots of history with different idles like approve, idle. also add credit cost breakdown

[NiamhRelevance]

Dont include voice sessions in invent features

[NiamhRelevance]

also remvoe subagent spawning

[NiamhRelevance]

remove feedback section

[NiamhRelevance]

analyse teh screenshots and find a suitable place on the page for them

[github-actions]

🎯 Vibe check

Reviewed: 1 file (1 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	Product term capitalization (Agent, Tool, Workforce, Trigger, Knowledge) is inconsistently applied throughout — roughly 10 instances where the terms appear lowercased while referring to Relevance AI product features. No banned words or British spellings found.
🟡	Technical clarity	7/10	Model name "Claude Opus 4.8" (line 213) doesn't match any known Claude model (latest is Opus 4.7). Entry point for creating a tool with Inventor described differently here vs. create-a-tool.mdx.
🟢	Non-technical clarity	9/10	Intro is clean and jargon-free with a good plain-English definition. Feature tabs are detailed and well-organized.
🟡	Structure	7/10	No closing CTA for a concept/overview page. One <CardGroup cols={1}> with non-navigable short suggestions that would work better as a bullet list.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Solid, well-structured overview page with detailed feature coverage across tabs — the content is genuinely useful. The main drag is inconsistent product term capitalization (Agent, Tool, Workforce, etc. repeatedly appear lowercased), which is a findable-and-fixable pattern once you know what to look for. The page also needs a closing CTA to give concept-page readers somewhere to go next.

🔧 Issues (4)

• build/invent/invent.mdx:11,15,29,32,35,58,83,175,372,377 — Product terms (Agent, Tool, Workforce, Trigger, Knowledge) are systematically lowercased when referring to Relevance AI product features. Worst offenders:

  • Line 29: "the tools it needs, and wires the triggers that fire them" → "the Tools it needs, and wires the Triggers that fire them"
  • Line 32: "Create new agents, tools, workforces, and triggers" → "Agents, Tools, Workforces, and Triggers"
  • Line 35: "tool vs. agent vs. workforce" → "Tool vs. Agent vs. Workforce"
  • Line 83: "tool, agent, workforce, knowledge, and home" → "Tool, Agent, Workforce, Knowledge, and home" (home is a generic page label, not a product name)
  • Line 175: "agents, tools, workforces, and triggers" → "Agents, Tools, Workforces, and Triggers"

• build/invent/invent.mdx:213 — "Claude Opus 4.8" does not match any known Claude model release; latest Opus is 4.7. Verify this model name is correct before publishing — if it's a placeholder, update it.

• build/invent/invent.mdx:153 — Entry point described as "Click the Build with Inventor button in the header." The sibling page build/tools/create-a-tool.mdx:22–23 describes this as clicking + New Tool then selecting Invent — different button label and flow for what appears to be the same action. One of these is out of date; align the UI instructions.

• build/invent/invent.mdx:203 — "your Invent session" — the product feature is "Inventor," not "Invent." Use "your Inventor session" for consistency with all other references on the page.

🧩 Component suggestions (2)

• build/invent/invent.mdx:302–312 — <CardGroup cols={1}> with 3 non-navigable, short suggestion items ("Build from a discussion", "Discuss best practices", "Improve reliability"). Per CLAUDE.md, CardGroups are for navigable or parallel-choice items with real content per card. These three are short descriptors without links; a plain bulleted list would be lighter and just as readable.

• build/invent/invent.mdx:354–360 — <CardGroup cols={2}> for Thumbs up / Thumbs down contains two items that are essentially just labels with one-sentence descriptions. A simple bold-text-with-description pattern (e.g. **Thumbs up** — The response was accurate and helpful.) or two-item bullet list would be less visually heavy for such short content.

🏗️ Page structure (1)

• build/invent/invent.mdx — No closing CTA. This is a concept/overview page and ends abruptly after the MCP comparison <Tip>. Readers who just learned what Inventor is need somewhere to go. A ## What's next? section linking to 2–3 logical follow-ons would close the loop — good candidates: /build/agents/build-your-agent/overview (build your first agent), /build/tools/create-a-tool (already referenced inline), and /build/agents/build-your-agent/evals (also linked inline and a natural follow-on for the quality/monitoring angle).

⚠️ Contradictions (1)

• build/invent/invent.mdx:148–155 vs build/tools/create-a-tool.mdx:22–23 — Two different flows are described for starting Inventor from the Tools page. invent.mdx says click Build with Inventor in the header. create-a-tool.mdx says click + New Tool then select Invent. These may be two different paths to the same result, but they use different button names with no indication that both routes exist — a user following one page's instructions won't recognize the UI described on the other.

🔋 Credit usage

Item	Count
Files reviewed	1
Context pages read	2
Total lines processed	~520

Files read: build/invent/invent.mdx (390 lines), integrations/mcp/mcp-server.mdx (332 lines — partially), build/tools/create-a-tool.mdx (102 lines)

[claude]

Changes made in response to review

Thanks for the detailed feedback, @NiamhRelevance. Here's what I've addressed:

Product name

• Renamed "Inventor" → "Invent" throughout the entire page (title, sidebarTitle, description, all body text)

Removed

• Voice sessions tab (removed from Invent features)
• Subagent spawning tab (removed from Invent features)
• Feedback section (removed from Invent features)
• The /invent route deprecation notice

Added

• Access section — Invent is available by default for all org admins; each project has a toggle to extend access to all users in that project
• Alarms and alerts — new Invent features accordion covering eval-based alerts: threshold, sensitivity, and "just ask" setup flow; includes the four eval screenshots you shared (monitor, create evals, fix failing evals, alarms config)
• What's next? — closing CTA linking to agent builder overview, create a tool, and evals
• FAQ section — moved the bug-logging note here ("Does Invent log bugs automatically?")

Restructured

• Invent features — changed from a <Tabs> block (was getting long) to an <AccordionGroup> so all feature names are visible at a glance and users only expand what they need
• Scheduled check-ins — replaced the mechanical step-by-step with your copy: covers how to ask, how the interval is suggested, and what happens when a check-in runs
• Model selection — added explanation that clicking a model shows its capabilities (text only vs. text and images); added model capabilities screenshot; fixed model tier list from cards to prose; fixed Claude Opus version 4.8 → 4.7; spelled out "large language model (LLM)" on first use
• Credit usage — added the task statuses screenshot (approve, idle, completed) with explanation of the history view

Cleaned up

• "What you can do" — removed voice and subagent cards, added Schedule check-ins and Set alarms cards, updated product term capitalization (Agents, Tools, Workforces, Triggers)
• About you — replaced 3-card CardGroup with a prose bullet list
• Suggested actions — replaced single-column CardGroup for general starting points with a bullet list
• Callout types: <Info> → <Warning> for memory deletion; <Info> → <Check> for conversation persistence

Images
The following image files need to be added to /images/ before the page renders correctly:

• invent-start-menu.png
• invent-model-capabilities.png
• invent-alarms.png
• invent-evals-monitor.png
• invent-create-evals.png
• invent-fix-failing-evals.png
• invent-task-statuses.png
• invent-history.png

(Existing images like auto-accept.png, invent-model-selector.png, about-you-invent.png, invent-credit-counter.png, and the location screenshots are unchanged.)

[github-actions]

🎯 Vibe check

Reviewed: 1 file (1 with issues, 0 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	7/10	Product term capitalization lapses for "Agent" and "Tool" scattered across card titles, tab titles, step titles, and body text.
🟢	Technical clarity	9/10	One minor alt text mismatch (status label wording). Otherwise specific, precise UI references throughout.
🟢	Non-technical clarity	9/10	No issues found.
🟢	Structure	9/10	One accordion uses bold text as sub-section dividers where headings would be correct. Everything else is well-layered.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5. Scores are a single overall judgment about the whole PR — not per file.

✨ Overall vibe: Solid, well-written concept page — clear definition, good layering, appropriate use of components throughout, and a proper CTA. The main thing dragging it down is inconsistent product term capitalization: "Agent" and "Tool" are correctly capped in most places but slip to lowercase in several card titles, tab titles, and step titles. A quick pass to align those will bring this to a clean state.

🔧 Issues (9)

• build/invent/invent.mdx:11 — "specialist agents" → "specialist Agents" (these are Relevance AI Agents, not a generic concept)
• build/invent/invent.mdx:55 — Card title "Debug failing agents" → "Debug failing Agents"
• build/invent/invent.mdx:99 — Tab title "While editing an agent" → "While editing an Agent"
• build/invent/invent.mdx:110 — Step title "Open or create an agent" → "Open or create an Agent"
• build/invent/invent.mdx:118 — Tab title "When creating an agent" → "When creating an Agent"
• build/invent/invent.mdx:132 — Tab title "When creating a tool" → "When creating a Tool"
• build/invent/invent.mdx:180 — Alt text says "approve, idle, and other statuses" but the body text on line 182 lists the statuses as "approved, idle, or completed" — align them (prefer the body text wording)
• build/invent/invent.mdx:347 — Card title "Build your first agent" → "Build your first Agent"
• build/invent/invent.mdx:350 — Card title "Create a tool" → "Create a Tool"

🧩 Component suggestions (1)

• build/invent/invent.mdx:232,236 — Inside the "Scheduled check-ins" accordion, **Scheduling a check-in** and **What happens when a check-in runs** are bold text acting as section sub-headers. CLAUDE.md says not to mix bold text with heading formats for section titles. Replace with #### headings: #### Scheduling a check-in and #### What happens when a check-in runs.

⚠️ Contradictions (0)

No contradictions found between this page and the linked pages checked (/build/agents/build-your-agent/evals, /build/tools/create-a-tool).

🔋 Credit usage

Item	Count
Files reviewed	1
Context pages read	2
Total lines processed	~477

Files read: build/invent/invent.mdx (357 lines), build/agents/build-your-agent/evals.mdx (60 lines, partial), build/tools/create-a-tool.mdx (60 lines, partial)

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	build/introduction.mdx has systematic product-term capitalization errors throughout — "agents", "tools", "workforces", "knowledge" appear lowercase in body text even though they're Relevance AI product features. Both files have individual accordion title mismatches.
🟡	Technical clarity	8/10	"evals" appears a dozen times in invent.mdx without ever being defined. The Alarms accordion mentions "threshold and sensitivity level" but doesn't explain what the threshold measures.
🟡	Non-technical clarity	7/10	"forward deployed agent" (invent.mdx:11) is unexplained jargon — the sentence defines what Invent does, but not what "forward deployed" means. "Evals" and "eval suite" are undefined on first use.
🟡	Structure	8/10	<Steps> in introduction.mdx wraps four parallel concepts (Agents, Tools, Workforces, Knowledge), not a sequential procedure — misuse of the component.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Both pages are well-written — clear, direct, and free of banned words and British spellings. The main drag is introduction.mdx's consistent failure to capitalize product terms in body text despite getting the titles right, which reads as inconsistent to anyone who knows the style guide. invent.mdx is otherwise solid, with good tone and structure; it just needs "evals" defined on first use.

🔧 Issues (11)

• build/introduction.mdx:8 — "agents, tools, workforces, and knowledge" → "Agents, Tools, Workforces, and Knowledge" (these are the Relevance AI product features, not generic terms)
• build/introduction.mdx:21 — "You create [agents]" → "You create [Agents]"
• build/introduction.mdx:22 — "use your tools and knowledge" → "use your Tools and Knowledge"
• build/introduction.mdx:24 — "You build [tools]...your agents can do" → "You build [Tools]...your Agents can do"
• build/introduction.mdx:27 — "multiple agents into a [workforce]" → "multiple Agents into a [Workforce]"
• build/introduction.mdx:30 — "You connect [knowledge] so agents have access" → "You connect [Knowledge] so Agents have access"
• build/introduction.mdx:49 — Four product terms lowercase in the same sentence: "agent's prompt" → "Agent's prompt"; "custom tools" → "custom Tools"; "knowledge sources" → "Knowledge sources"; "in a workforce" → "in a Workforce"
• build/invent/invent.mdx:69 — Accordion title="About you" → title="About You". Every mention in the body text (lines 72, 83) writes it as About You, which is the actual UI element name.
• build/invent/invent.mdx:129 — Accordion title="Auto-accept and tool approval" → title="Auto-accept and Tool approval". The body refers to tool runs from the Invent Tool feature, not a generic tool concept.
• build/invent/invent.mdx:11 — "forward deployed agent" is undefined jargon for a non-technical reader. Consider cutting "forward deployed" or glossing it in a parenthetical: "Invent is an agent built into Relevance AI that works alongside you (a forward-deployed agent — one that's embedded in your workflow rather than called separately)."
• build/invent/invent.mdx:27 — "evals" appears here for the first time with no definition. A short gloss on first use would help: "evals (automated tests that check whether an Agent's output meets your quality criteria)". The term appears ~12 times in the file.

🧩 Component suggestions (3)

• build/introduction.mdx:19–32 — <Steps> is used for four parallel concepts (Agents — the workers, Tools — the actions, Workforces — the team, Knowledge — the context). These are not sequential steps; they're equivalent building blocks. Replace with <CardGroup cols={2}>, where each card carries the same icon, title, and description it has now, and links to the matching create page already used in the body. This gives readers navigable entry points and is what CLAUDE.md recommends for equal, parallel feature concepts.

• build/invent/invent.mdx:20–31 — The four <Accordion> components in "What can Invent do?" use a description prop (description="Learn the work, systems, and judgment calls"). This prop isn't present on any Accordion elsewhere in the repo — verify it renders the subtitle as intended in Mintlify rather than being silently dropped.

• build/invent/invent.mdx:105–118 — <CardGroup> inside "Suggested actions" holds four cards (Add test coverage, Add production monitoring, etc.) that don't link anywhere and each contain a single short sentence. Per CLAUDE.md, CardGroup is inappropriate for non-navigable descriptor lists. A plain bulleted list — or the prose already used for the non-Agent-context suggestions immediately below — would be simpler and more consistent within the same accordion.

🏗️ Page structure (1)

• build/invent/invent.mdx:247–249 — "Start with one process" works as closing guidance but has no links. For a concept page, this is where readers need a path forward. Consider adding 1–2 links, for example /build/agents/create-an-agent (to build the first Agent Invent creates) and /build/invent/invent is already the page — something like a link to a related tutorial or the Workforce page would give readers a concrete next action rather than leaving them to navigate the sidebar.

✅ Clean files (0)

Both MDX files had issues.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~735

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (250 lines), integrations/mcp/mcp-server.mdx (332 lines), integrations/mcp/mcp-client.mdx (103 lines)

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	7/10	Five capitalization misses — product terms (Agent, Tool, Workforce) are lowercase in introduction.mdx lines 9, 41, 49 and once each in invent.mdx accordion title and body (lines 120, 125).
🟡	Technical clarity	8/10	"Forward deployed agent" is dropped with no definition on first use. Otherwise UI references are specific and the content is accurate.
🟡	Non-technical clarity	7/10	Same jargon issue — "forward deployed agent" will land flat for a business reader. Rest of the page is unusually readable for a feature doc this complex.
🟡	Structure	7/10	invent.mdx FAQ section copy-pastes verbatim content already covered in the Access controls section above. Four accordion items are exact duplicates. introduction.mdx uses <Steps> for parallel concepts rather than sequential steps.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The Invent page is genuinely good — well-organized, thorough, and one of the clearer feature pages in the repo. The main drag is a structural one: the FAQ section duplicates the Access controls content word-for-word, which creates bloat without adding value. Fix the duplication and sort the capitalization pattern and this is a clean merge.

🔧 Issues (6)

• build/introduction.mdx:9 — "agents, tools, workforces, and knowledge" → "Agents, Tools, Workforces, and Knowledge". These are Relevance AI product features; the image alt-text on line 15 of the same file already capitalizes them correctly.
• build/introduction.mdx:41 — "Invent builds the agents to do it" → "the Agents to do it"
• build/introduction.mdx:49 — Three instances in one sentence: "an agent's prompt" → "an Agent's prompt", "custom tools" → "custom Tools", "a workforce" → "a Workforce"
• build/invent/invent.mdx:120 — Accordion title "Auto-accept and tool approval" → "Auto-accept and Tool approval" (refers to Relevance AI Tool executions, not tools generically)
• build/invent/invent.mdx:125 — "tool runs that require your approval" → "Tool runs" (same reason as above)
• build/invent/invent.mdx:11 — "forward deployed agent" is undefined jargon on first use. Add a one-clause definition inline — e.g. "a specialist AI agent built into the Relevance AI platform" — or explain what "forward deployed" means for a reader who hasn't heard the term before.

🧩 Component suggestions (1)

• build/introduction.mdx:19–32 — <Steps> implies a sequential numbered procedure, but Agents, Tools, Workforces, and Knowledge are parallel building blocks, not ordered steps. A <CardGroup cols={2}> (matching the "Ways to create" section directly below) would be more semantically correct. Each item already has an icon and a short description — it maps cleanly to a Card.

🏗️ Page structure (3)

• build/invent/invent.mdx:56 vs 244–246 — The "Availability" section and the "Who can use Invent?" FAQ answer are identical sentences. The FAQ should either be removed (the answer is already visible above the fold) or replaced with a cross-reference: "See Availability."
• build/invent/invent.mdx:60–72 vs 248–258 — The three Enterprise security control accordions (Project isolation, Integration reach, Data egress) are repeated verbatim in the FAQ section as three separate FAQ entries. Remove the FAQ duplicates and link to #enterprise-security-controls from any FAQ question that asks about them.
• build/invent/invent.mdx — No closing CTA. This is a concept/feature overview page, so a brief ## What's next? belongs at the end. Two natural targets already in the repo: /build/introduction (to see where Invent sits in the broader build platform) and the /get-started/quickstart page (if one exists) or whichever page walks through the first Invent session end-to-end.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~744

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (272 lines), integrations/mcp/mcp-server.mdx (362 lines), integrations/mcp/mcp-client.mdx (60 lines, partial)

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟢	Consistency	8/10	Two "tool"/"Tool" capitalization misses in invent.mdx (accordion title + body copy). No banned words, no British spellings, headings are otherwise clean.
🟢	Technical clarity	9/10	Four images referenced in invent.mdx are not included in this PR — they may be pre-existing, but worth confirming they exist in the repo.
🟢	Non-technical clarity	9/10	Well-written throughout. Plain language, no jargon dumps. The credit warning is appropriately prominent.
🟡	Structure	7/10	Three enterprise security FAQ entries duplicate the "Access controls" section verbatim — no condensing, no link-back. "Who can use Invent?" FAQ does the same to the "Availability" sub-section. invent.mdx also lacks a closing CTA.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5. Scores are a single overall judgment across the whole PR.

✨ Overall vibe: Solid, well-written docs for a feature-rich product area — good plain language, correct structure of nested tabs/accordions, and useful comparisons. The main drag is structural: the FAQ section copies the enterprise security controls and availability content verbatim from the sections above rather than summarizing or linking to them, adding ~200 words of repeated content with no added value. Fix that and this is in very good shape.

🔧 Issues (6)

• build/invent/invent.mdx:125 — Accordion title "Auto-accept and tool approval" — "tool" should be "Tool". The body text makes clear these are Relevance AI Tools (the product feature) being queued for approval, not generic tool-use.

• build/invent/invent.mdx:130 — "tool runs that require your approval appear in the approval queue" — same: "Tool runs" (Relevance AI product feature).

• build/invent/invent.mdx:52–72 vs 249–263 — The body text in the three enterprise FAQ accordions ("How does project isolation work for Enterprise customers?", "Which integrations can Invent access in an Enterprise project?", "Where can Enterprise Invent data be sent?") is copied word-for-word from the "Enterprise security controls" <AccordionGroup> above. CLAUDE.md: "Don't duplicate content across pages — link to the canonical source instead." The FAQ entries should either summarize in one sentence and link to the section above, or the top-level accordions should be removed in favour of the FAQ.

• build/invent/invent.mdx:56 vs 249–250 — The FAQ entry "Who can use Invent?" repeats the "Availability" sub-section verbatim. Same fix: summarize or link.

• build/invent/invent.mdx:95 — <img src="/images/about-you-invent.png" ...> — this image is not in the PR's changed files. Confirm it exists in the repo already.

• build/invent/invent.mdx:113,126,147 — Three more images referenced but not in the PR: invent-model-selector.png, auto-accept.png, invent-credit-counter.png. Likely pre-existing, but verify.

🧩 Component suggestions (1)

• build/introduction.mdx:19–32 — <Steps> is used to list Agents, Tools, Workforces, and Knowledge. These are four parallel, independent components — not a sequence the user performs in order. A <CardGroup cols={2}> with icons (the icons are already on each <Step>) would signal "four equal options" rather than "four ordered steps". The existing <Steps> for sequential procedures like "Open Invent → Click Settings" in invent.mdx shows the right use case.

🏗️ Page structure (1)

• build/invent/invent.mdx — Page ends with the FAQ accordions but has no closing CTA. This is a concept/overview page, so readers who just learned what Invent is and how it works need somewhere to go next. Suggest adding a ## What's next? section with links to /build/agents/create-an-agent, /build/tools/create-a-tool, and /build/workforces/create-a-workforce — the core building blocks Invent creates for you.

✅ Clean files (0)

Both changed MDX files have issues. No clean files to list.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~514

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (277 lines), integrations/mcp/mcp-server.mdx (362 lines), integrations/mcp/mcp-client.mdx (103 lines)

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	introduction.mdx uses lowercase "agents", "tools", "workforces", "knowledge" in prose and the description frontmatter — but capitalizes them in Step titles and image alt text within the same file. invent.mdx FAQ duplicates the Access controls section verbatim across 3 entries.
🟡	Technical clarity	7/10	Auto-accept is mentioned in the FAQ but never documented (no steps, no image, no explanation of where to find it). Three images added in this PR (auto-accept.png, invent-credit-counter.png, invent-model-selector.png) are not referenced in any MDX file — they will not render.
🟢	Non-technical clarity	9/10	Invent is explained clearly in plain language. The "About You", Memories, and Bug reporting accordions are appropriately scoped. The MCP vs Invent comparison section is genuinely useful.
🟡	Structure	7/10	The FAQ in invent.mdx duplicates content from Access controls instead of linking to it. No closing CTA on a comprehensive concept page.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The Invent page is well-organized and clearly written — the feature breakdown, accordion grouping, and MCP comparison are all solid. Two things need fixing before merge: a handful of product term capitalization misses in introduction.mdx, and three images uploaded in this PR that aren't referenced anywhere and will never appear on the page.

🔧 Issues (11)

• build/introduction.mdx:4 — description frontmatter: "agents, tools, and workforces" → "Agents, Tools, and Workforces" (product feature names)
• build/introduction.mdx:9 — "agents, tools, workforces, and knowledge" → "Agents, Tools, Workforces, and Knowledge" (same paragraph capitalizes them in the alt text on line 15)
• build/introduction.mdx:11 — "interact with your agents directly" → "interact with your Agents directly"
• build/introduction.mdx:22 — "You create [agents]..." → "You create [Agents]..." (the link destination makes clear these are the product feature)
• build/introduction.mdx:24 — "what your agents can do" → "what your Agents can do"; "[tools]" link text → "[Tools]"
• build/introduction.mdx:26 — "multiple agents into a [workforce]" → "multiple Agents into a [Workforce]"
• build/introduction.mdx:29 — "so agents have access" → "so Agents have access"; "[knowledge]" link text → "[Knowledge]"
• build/introduction.mdx:49 — "an agent's prompt" → "an Agent's prompt"
• build/invent/invent.mdx:270–280 — Three FAQ entries ("How does project isolation work…", "Which integrations can Invent access…", "Where can Enterprise Invent data be sent?") copy the Access controls → Enterprise security controls accordions word for word. This duplicates content that already lives a few scrolls above on the same page. Link the FAQ answers to the #enterprise-security-controls section instead of repeating the text.
• build/invent/invent.mdx:283 — FAQ answer mentions "Auto-accept" as a feature with behavior ("Invent apply changes without pausing for each approval") but there is no accordion, step, or section explaining how to enable or disable it. images/auto-accept.png was added in this PR but is not embedded anywhere.
• Orphaned images — images/auto-accept.png, images/invent-credit-counter.png, and images/invent-model-selector.png are all added in this PR but not referenced in either changed MDX file. They will not appear anywhere on the docs site.

🏗️ Page structure (1)

• build/invent/invent.mdx — No closing CTA. This is a concept + feature overview page; ending on FAQ leaves readers with no forward momentum. A short ## What's next? section pointing to /build/agents/create-an-agent and /build/agents/build-your-agent/evals would round it off.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2 (evals.mdx, mcp-server.mdx partial)
Total lines processed	~841

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (294 lines), build/agents/build-your-agent/evals.mdx (447 lines), integrations/mcp/mcp-server.mdx (50 lines, partial)

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	6/10	introduction.mdx uses lowercase for product names (Agent, Tool, Workforce, Knowledge) in body text throughout — 9+ instances across a 50-line file, even though the Step titles above capitalize them correctly. invent.mdx handles product term capitalization well.
🟡	Technical clarity	8/10	invent.mdx "Manage Evals with Invent" section omits the progressive rollout restriction noted prominently on the linked Evals page (Enterprise-only). Enterprise security content is duplicated verbatim between the reference section and FAQ, making it harder to maintain.
🟡	Non-technical clarity	8/10	"Forward deployed agent" in both the frontmatter description and the opening sentence is unexplained jargon — Relevance AI's target audience includes non-technical business users who won't know what this means. Everything else reads clearly.
🟡	Structure	7/10	invent.mdx has no closing CTA for what is a concept overview page. Enterprise security content appears word-for-word in both the "Enterprise security controls" accordion group and the FAQ section — significant duplication with no cross-referencing.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5. Scores are a single overall judgment about the whole PR — not per file. The dropdowns below contain the line-by-line specifics.

✨ Overall vibe: The invent.mdx is thorough and well-constructed — good use of tabs, accordions, steps, and callouts, with clean product terminology and solid UI references throughout. The main drag is a repetitive product term capitalization failure in introduction.mdx and a content duplication issue in the invent page that makes the FAQ section feel padded rather than useful. Fix the capitalization, fold the duplicate enterprise security content, add a CTA, and this is in good shape.

🔧 Issues (9)

• build/introduction.mdx:4 — frontmatter description: "agents, tools, workforces" → "Agents, Tools, Workforces" (Relevance AI product names)
• build/introduction.mdx:9 — "agents, tools, workforces, and knowledge" → "Agents, Tools, Workforces, and Knowledge"
• build/introduction.mdx:21 — "You create [agents]" → "You create [Agents]"
• build/introduction.mdx:24 — "[tools]" in body text → "Tools"
• build/introduction.mdx:27 — "[workforce]" in body text → "Workforce"
• build/introduction.mdx:30 — "[knowledge]" in body text → "Knowledge"
• build/introduction.mdx:36 — "When creating agents and tools" → "When creating Agents and Tools"
• build/introduction.mdx:49 — "an agent's prompt" → "an Agent's prompt"; "custom tools" → "custom Tools"; "knowledge sources" → "Knowledge sources"; "workforce" → "Workforce"
• build/invent/invent.mdx:5,11 — "forward deployed agent" is unexplained jargon in both the frontmatter description and the opening sentence. Relevance AI's audience includes non-technical business users. Consider plain-English phrasing or a brief inline gloss on first use.

🧩 Component suggestions (2)

• build/invent/invent.mdx:166–168 — <Note> warns that you need an alert channel configured before a scheduled check-in will notify you of anything. Without a channel, check-ins run silently with no output — that's a prerequisite failure, not a helpful tip. <Warning> fits better here.
• build/invent/invent.mdx:7,36 — Two images use raw <img> tags while the rest of the repo (including the linked evals page) uses Markdown image syntax (![alt](/path/image.png)). Both render in Mintlify, but <img> bypasses any image optimization or lazy-loading Mintlify applies to Markdown images. Worth normalizing for consistency.

🏗️ Page structure (2)

• build/invent/invent.mdx — No closing CTA. This is a full concept + overview page; after the FAQs, readers have nowhere obvious to go next. Good candidates: /build/agents/create-an-agent (start building what Invent will manage), /build/agents/build-your-agent/evals (set up quality checks), /build/workforces/create-a-workforce (wire Agents together).
• build/invent/invent.mdx:60–72 vs build/invent/invent.mdx:306–316 — The three enterprise security accordions ("Project isolation", "Integration reach", "Data egress") are reproduced word-for-word in the FAQ section. Pick one canonical home and link from the other, or collapse the FAQs into the reference section with anchor links.

⚠️ Contradictions (1)

• build/invent/invent.mdx (Manage Evals section, lines 199–272) vs build/agents/build-your-agent/evals.mdx:7–9 — The evals page has a prominent rollout notice: "Evals is rolling out progressively, starting with Enterprise customers. If you don't see this feature in your account yet, reach out to your account manager to discuss access." The invent.mdx "Manage Evals with Invent" section presents Evals as fully available to any Invent user with no such caveat. A non-Enterprise user following the Create Evals steps will hit a dead end without understanding why.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~1,189

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (330 lines), build/agents/build-your-agent/evals.mdx (447 lines), integrations/mcp/mcp-server.mdx (362 lines)

[github-actions]

🎯 Vibe check

Reviewed: 2 files (1 with issues, 1 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	8/10	invent.mdx: the word "Powerful" appears in a UI-label context (borderline); "Performance dashboard" and "Monitor dashboard" are used for what appears to be the same UI element across pages.
🟡	Technical clarity	8/10	invent.mdx: the "Configure an alarm" tab opens with instructions before defining what an alarm is; anchor links to #alerts inside a <Tab> may not resolve correctly.
🟢	Non-technical clarity	9/10	Plain language throughout, no jargon dumps, good definition-before-instructions pattern in almost every section.
🟡	Structure	7/10	invent.mdx: three enterprise security accordions are copied verbatim into the FAQ section; no closing CTA on a concept page.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The Invent page is thorough and well-written — it earns trust by being specific rather than vague, and the tab/accordion structure tames a genuinely large feature surface. The main structural drag is the word-for-word duplication of enterprise security content between the "Access controls" section and three FAQ items, which should be collapsed into a single canonical location.

🔧 Issues (3)

• build/invent/invent.mdx:101–112 vs 346–356 — The three enterprise security accordions ("Project isolation", "Integration reach", "Data egress") are copied verbatim into the FAQ. CLAUDE.md: "No duplicate content — link to canonical sources." Remove the FAQ duplicates and replace each with a single sentence pointing up to the Enterprise security controls section.

• build/invent/invent.mdx:162 — **Powerful** appears in prose: "Labels such as Fast, Balanced, Powerful, and Thorough…". It's quoting a UI label, not describing Invent — but Powerful is on the banned-words list. Consider rephrasing to make the quoting intent unambiguous, e.g. "Labels — Fast, Balanced, Powerful, and Thorough — summarize…" with a note like "(as labelled in the dropdown)".

• build/invent/invent.mdx:307–309 — The <Note> callout contains bold text **Evaluate → Monitor**. CLAUDE.md says callouts should have no bold labels inside. Rephrase as plain text: "Alarms are configured in an Agent or Workforce's Evaluate → Monitor area, not in Invent settings."

🏗️ Page structure (3)

• build/invent/invent.mdx:299–311 — The "Configure an alarm" tab opens with instructions for enabling Auto-diagnosis, then a Note about where alarms live, then finally a one-line definition of what an alarm is. Flip this: lead with the definition ("Alarms evaluate scores collected by a Performance dashboard and notify you when a score or Check crosses the condition you set"), then point to where they're configured (Evaluate → Monitor), then cover Auto-diagnosis.

• build/invent/invent.mdx — No closing CTA. This is a concept/feature overview page; per CLAUDE.md it should end with a CTA. A natural close would be a ## What's next? section pointing to /build/agents/create-an-agent (to start building the Agent Invent will manage) and /build/agents/build-your-agent/evals (to set up the quality bar Invent monitors).

• build/invent/invent.mdx:207 and 304 — Both [Alerts](/build/invent/invent#alerts) anchor links target an accordion nested inside a <Tab>. Mintlify may not resolve nested-tab anchors reliably. Verify these render as working deep-links in the preview; if not, link to the section heading instead (#access-controls) or restructure Alerts as a top-level section.

⚠️ Contradictions (1)

• build/invent/invent.mdx:272,310 calls the monitoring UI element "Performance dashboard". build/agents/build-your-agent/evals.mdx:300–305 calls the same element a "Monitor dashboard" (or just "dashboard"). Both pages point readers to the same feature under Evaluate → Monitor. Pick one name and align them — the evals page is the canonical source for this feature, so "Monitor dashboard" is the safer anchor term.

✅ Clean files (1)

build/introduction.mdx — Sentence case clean, no banned words, appropriate use of <Steps> and <CardGroup>, CTA present.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~946

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (369 lines), build/agents/build-your-agent/evals.mdx (447 lines), integrations/mcp/mcp-server.mdx (80 lines, partial)

[alex-duckmanton]

Some feedback:

Invent reads outcomes from the systems where the work lands, not only from Agent logs, so it can connect Agent activity to measures such as response time, qualified leads, meetings booked, or cases resolved. It can also identify recurring manual work and recommend additional Agents worth building.

Can it do this now, or is this something coming in the future?

Bottom-right button

Rename this to "Floating button" and mention it's available on every page

[github-actions]

🎯 Vibe check

Reviewed: 2 files (1 with issues, 1 clean)

Scores

	Dimension	Score	What's holding it back
🟢	Consistency	9/10	"forward deployed" missing hyphen (2 occurrences in invent.mdx — frontmatter and body). Everything else is clean: sentence case throughout, no banned words in author-written copy, American English.
🟡	Technical clarity	7/10	Two #alerts anchor links target an Accordion title, not a proper heading — likely broken. No rollout caveat in the Invent+Evals section despite the Evals page displaying a prominent Enterprise-only banner. Three images in the PR diff are not referenced in any changed MDX file.
🟢	Non-technical clarity	9/10	Well-defined throughout. The "Configure an alarm" tab puts the general definition of what alarms do after the auto-diagnosis option, which slightly inverts the expected setup → feature order.
🟡	Structure	7/10	Enterprise security controls content is duplicated verbatim in both the body AccordionGroup (lines 100–112) and the FAQ section (lines 340–354) of invent.mdx. The "Configure an alarm" tab ordering is also inverted.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: invent.mdx is a solid, well-componentized page — sentence case is consistent, callout types are used correctly, and the prose is specific and non-marketing. The main things to fix before shipping are the duplicate enterprise security content (clean that up or link from FAQ to the body section), the two potentially broken #alerts anchors, and the missing Evals rollout note that the Evals page itself prominently displays.

🔧 Issues (5)

• build/invent/invent.mdx:5 (frontmatter description) — "forward deployed agent" → "forward-deployed agent". Compound adjective before a noun needs a hyphen.
• build/invent/invent.mdx:11 — "Invent is a forward deployed agent" → "forward-deployed agent". Same issue in the body.
• build/invent/invent.mdx:204 — [Alerts](/build/invent/invent#alerts) — #alerts targets an Accordion title, not a proper heading. CLAUDE.md: "the target must be a proper heading." Mintlify does not generate heading anchors from <Accordion title="...">. Either add a ## Alerts heading above the accordion or remove the anchor.
• build/invent/invent.mdx:302 — [Invent alert channels](/build/invent/invent#alerts) — same broken anchor as above.
• PR-level — images/auto-accept.png, images/invent-credit-counter.png, and images/invent-model-selector.png are in the PR diff but not referenced in any changed MDX file. Confirm they're used in another existing page, or remove them if orphaned.

🏗️ Page structure (3)

• build/invent/invent.mdx:100–112 and build/invent/invent.mdx:340–354 — The three Enterprise security controls accordions ("Project isolation", "Integration reach", "Data egress") have their full body text duplicated verbatim in the FAQ section. The FAQ answers are word-for-word identical to the accordion content above. Consolidate: replace the three FAQ entries with a single entry that links to the ## Enterprise security controls section, or drop the body accordions and keep it only in the FAQ.
• build/invent/invent.mdx:298–309 — "Configure an alarm" tab: the general description of what alarms do ("Alarms evaluate the scores collected by a Performance dashboard and notify you when an overall score or individual Check crosses the condition you set") appears on line 307–309, after the auto-diagnosis feature and the Note about where alarms are configured. Flip the order: define what alarms do first, then introduce auto-diagnosis, then the placement Note.
• build/invent/invent.mdx:237–310 — The "Manage Evals with Invent" section makes no mention that Evals is currently an Enterprise-only rollout. The canonical evals.mdx page opens with a prominent Info banner: "Evals is rolling out progressively, starting with Enterprise customers." Users who land on the Invent page and see the Evals tabs will have no indication the feature may not be in their account. Add a brief note here, or link to the Evals page's rollout status.

⚠️ Contradictions (1)

• build/invent/invent.mdx:237–310 vs build/agents/build-your-agent/evals.mdx:7–9 — The Invent page presents Evals features (Create, Review, Fix, Alarms) without any availability qualifier. The Evals page leads with "Evals is rolling out progressively, starting with Enterprise customers. If you don't see this feature in your account yet, reach out to your account manager." These two pages make inconsistent implied promises about who can use this feature.

✅ Clean files (1)

build/introduction.mdx

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	2
Total lines processed	~915

Files read: build/introduction.mdx (50 lines), build/invent/invent.mdx (368 lines), build/agents/build-your-agent/evals.mdx (447 lines), integrations/mcp/mcp-server.mdx (~50 lines, partial)

[github-actions]

🎯 Vibe check

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

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	8/10	invent.mdx uses "Performance dashboard" while the canonical evals page uses "Monitor dashboard" throughout.
🟡	Technical clarity	7/10	Two anchor links target an Accordion title (#alerts) rather than a proper heading — these may silently fail. Terminology mismatch on dashboard naming.
🟢	Non-technical clarity	9/10	Plain-language definitions, good progressive disclosure with accordions, clear approval-controls explanation. No jargon dumps.
🟡	Structure	7/10	<Steps> used for non-sequential components in introduction.mdx. Enterprise security accordion content is repeated verbatim in the FAQ on the same page. No closing CTA on the invent concept page.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Both pages are well-written — clear, appropriately technical, no banned words, and capitalization is clean throughout. The main drag is structural: the enterprise security content is duplicated word-for-word between the accordion section and the FAQ, and a <Steps> component is used in introduction.mdx for parallel components rather than a sequential procedure.

🔧 Issues (3)

• build/invent/invent.mdx:260,293 — "Performance dashboard" is used twice but evals.mdx consistently calls these "Monitor dashboards" (the tab in the Evaluate area is named Monitor, not Performance). Pick one term and align with the canonical Evals page.

• build/invent/invent.mdx:195 — [Alerts](/build/invent/invent#alerts) targets #alerts, which resolves to an <Accordion title="Alerts"> inside a <Tab>, not a proper heading. Mintlify does not guarantee anchor generation for accordion titles. Same issue at line 293: [Invent alert channels](/build/invent/invent#alerts). If the accordion doesn't produce a working anchor, both links silently 404. Either promote "Alerts" to a ### heading or remove the anchor and link just to /build/invent/invent.

• build/invent/invent.mdx:90-102 and 334-344 — The three Enterprise security accordions ("Project isolation", "Integration reach", "Data egress") are pasted verbatim into the FAQ section. CLAUDE.md: "Don't duplicate content across pages — link to the canonical source instead." The same principle applies within a page. The FAQ entries should link to the accordion sections above (once the anchor issue above is fixed), or one of the two sets should be removed.

🧩 Component suggestions (1)

• build/introduction.mdx:31–44 — <Steps> is used to present Agents, Tools, Workforces, and Knowledge as if they're a sequential procedure, but they're parallel, co-equal components. CLAUDE.md says <Steps> is for sequential procedures. Swap for <CardGroup cols={2}> with icons — each card gets a title, icon, and one-line description with a link. This matches the CardGroup already used for "Choose how to build" a few lines later.

🏗️ Page structure (1)

• build/invent/invent.mdx — The page ends with the FAQ, no closing CTA. As a concept/overview page, readers who've worked through it need somewhere to go. Suggest adding a ## What's next? section after the FAQ with two or three links, for example: Build an agent, Evals, and MCP server.

⚠️ Contradictions (1)

• build/invent/invent.mdx:260,293 calls these "Performance dashboard"; build/agents/build-your-agent/evals.mdx:303,315 calls them "Monitor dashboard" or "Monitor dashboards" everywhere. Both pages describe the same UI area (the Monitor tab under Evaluate). Use "Monitor dashboard" to match the Evals page and the actual UI label.

🔋 Credit usage

Item	Count
Files reviewed	2
Context pages read	1
Total lines processed	~805

Files read: build/introduction.mdx (62 lines), build/invent/invent.mdx (358 lines), build/agents/build-your-agent/evals.mdx (447 lines)

[github-actions]

🎯 Vibe check

Reviewed: 3 files (2 with issues, 1 clean)

Scores

	Dimension	Score	What's holding it back
🟡	Consistency	7/10	checks lowercase on invent.mdx:27 while the same file uses Checks from line 229 onward; lowercase product terms in Steps body (introduction.mdx:33–42) while Step titles capitalize them; Knowledge missing from the Invent components list (invent.mdx:13) but present in both introduction.mdx descriptions.
🟡	Technical clarity	7/10	Two anchor links (#alerts) target an accordion title, not a heading — these links won't resolve per CLAUDE.md's rule that anchor targets must be proper headings. "Forward deployed agent" dropped in frontmatter and first sentence without definition.
🟢	Non-technical clarity	9/10	"Forward deployed agent" is the one unexplained term — everything else is well-defined, jargon-light, and logically ordered.
🟡	Structure	7/10	Enterprise security controls content (invent.mdx:91–102) is copied verbatim into three FAQ answers (invent.mdx:334–344); violates the no-duplication rule. No closing CTA on the Invent page.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Solid, well-organized introduction of a major feature — the writing is clear, the feature coverage is thorough, and the MCP vs Invent comparison is a useful clarification that will cut down support questions. The main problems are two broken anchor links (both target #alerts, which resolves to nothing because it's an accordion title, not a heading), verbatim content duplication between the security section and the FAQ, and a handful of capitalization inconsistencies.

🔧 Issues (6)

• build/invent/invent.mdx:27 — "checks" → "Checks". "Checks" is a product feature, capitalized consistently from line 229 onward in this file and throughout evals.mdx. The card body for "Set and maintain the quality bar" is the only place it's lowercase.

• build/invent/invent.mdx:195 — [Alerts](/build/invent/invent#alerts) anchor won't resolve. CLAUDE.md: "the target must be a proper heading." Alerts on this page is an <Accordion title="...">, not a heading — Mintlify doesn't generate anchor IDs from accordion titles. Either add a ## Alerts heading above the accordion or drop the anchor and link to /build/invent/invent alone.

• build/invent/invent.mdx:292 — Same #alerts issue in [Invent alert channels](/build/invent/invent#alerts).

• build/invent/invent.mdx:91–102 and 334–344 — Three FAQ answers ("How does project isolation work", "Which integrations can Invent access", "Where can Enterprise Invent data be sent") are word-for-word copies of the Project isolation, Integration reach, and Data egress accordions above. CLAUDE.md: "Don't duplicate content — link to the canonical source instead." Replace the FAQ answers with one sentence and a link: e.g. "See Enterprise security controls above."

• build/introduction.mdx:33, 36, 39, 42 — "agents", "tools", "knowledge", "workforce" are lowercase in the Steps body while the Step titles capitalize the same terms ("Agents — the workers", "Tools — the actions", etc.) and all four link to Relevance AI product pages. Inconsistent within the same component; should be Agents, Tools, Knowledge, Workforces.

• build/invent/invent.mdx:13 — "Agents, Tools, Triggers, and Workforces" omits Knowledge. Both descriptions of Invent in introduction.mdx (lines 21 and 52) include Knowledge. Align the list.

🧩 Component suggestions (1)

• build/invent/invent.mdx:82–84 — The Note in the Tool permissions tab covers three separate things: (1) these settings control approval prompts, not access; (2) Invent still uses existing project permissions; (3) each user manages their own settings and org/project admins can't use this screen to control others. The third point is the non-obvious gotcha — it's buried as the last sentence of a Note. Consider keeping only point 1 as the Note and surfacing point 3 as a standalone <Warning>: "Organization and project admins cannot use this screen to set approval rules for other users. Each user manages their own preferences per project."

🏗️ Page structure (1)

• build/invent/invent.mdx — No closing CTA after the FAQ. This is a concept/overview page; per the content standards, it should close with a path forward. Natural next steps: "Start building your first agent with Invent" → /build/agents/create-an-agent, or "Jump in with the quick start guide" → /get-started/quick-start-guide.

✅ Clean files (1)

docs.json — Navigation change is valid: Invent group placed before Agents in the Build tab, consistent with the introduction page ordering. No orphaned pages; the single-page Invent group is correct.

🔋 Credit usage

Item	Count
Files reviewed	3 (2 mdx + docs.json)
Context pages read	2 (evals.mdx, mcp-server.mdx)
Total lines processed	~2,365

Files read: build/introduction.mdx (62 lines), build/invent/invent.mdx (357 lines), docs.json (1,137 lines), build/agents/build-your-agent/evals.mdx (447 lines), integrations/mcp/mcp-server.mdx (362 lines)

[jordanc-relevanceai]

incredible