changelog: Tool Argument Matching (Evals) + web search & URL fetching (Invent) — July 15

[jordanc-relevanceai]

Collates two auto-drafted July 15, 2026 changelog entries into a single <Update> block, since both landed on the same date.

Supersedes:

• changelog: Tool Argument Matching Check Type for Evals #727 — Tool Argument Matching checks in Evals
• changelog: Web Search and URL Content Fetching in Invent #728 — Web search and URL content fetching in Invent

Content from both source PRs is preserved nearly verbatim, nested under two ### subheadings within one July 15 entry.

Notes:

• changelog: Tool Argument Matching Check Type for Evals #727 rollout verified: flag evals-toolusage-argument-matcher is at 100% with no conditions, and the check exists in relevance-api-node. Reachable by all users.
• changelog: Web Search and URL Content Fetching in Invent #728 (Invent web search) rollout not yet verified — worth a fact-check before merge.

🤖 Generated with Claude Code

[jordanc-relevanceai]

Closing the collated entry — keeping #727 (Evals) open on its own instead. #728 stays closed.

[mintlify]

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

Project	Status	Preview	Updated (UTC)
relevanceai	🔴 Failed	–	Jul 15, 2026, 6:24 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	3/10	15 banned-word hits across older entries (powerful ×11, seamlessly ×3, leverage ×1, unlock ×1, effortlessly ×1); 120 emoji-arrow instances (➡️); 22 bold-as-headings in title case; "Invent" vs "Inventor" naming mismatch in the new July 15 content.
🟡	Technical clarity	7/10	New content is specific and well-grounded (named operators, concrete to/subject example, exact UI paths). One actionable gap: the Evals docs don't yet document Tool Argument Matching, so the "See Evals documentation for full details" pointer is a dead end.
🔴	Non-technical clarity	5/10	New July 15 entry: clean. Pre-May 2026 entries: marketing copy with exclamatory closings ("Start syncing today!", "Your AI workforce just got a whole lot more collaborative!"), missing plain-English definitions.
🔴	Structure	4/10	New July 15 entry uses proper ##/### headings. Every entry from March 2026 and earlier uses **Bold Title Case** in place of a heading, creating a jarring inconsistency through roughly half the file.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: The July 15 entry being added by this PR is one of the better-written changelog entries in the file — technical, specific, no marketing fluff, proper heading hierarchy. The core issue to resolve before merge is that evals.mdx doesn't document Tool Argument Matching yet, so the "full details" link is a dead end for readers. The rest of the file's problems are pre-existing debt worth tracking separately.

🔧 Issues (16)

Naming mismatch (new content)

• changelog.mdx:24,26,28,30 — Uses "Invent" as the product noun ("Invent can now search the web…", "the standard Invent interface", "open Invent and paste a URL") but the linked page at /build/invent/invent titles the product "Inventor" and uses that name throughout. Pick one and apply it consistently across both files.

Banned words (all pre-existing)

• changelog.mdx:110 — powerful ("the same powerful auto-sync capabilities")
• changelog.mdx:236 — powerful ("using our powerful built-in search engine")
• changelog.mdx:289 — seamlessly ("You can now seamlessly integrate")
• changelog.mdx:405 — Unlock ("Unlock enterprise automation for Microsoft 365 teams")
• changelog.mdx:407 — seamlessly ("Integrate seamlessly with your existing workflows")
• changelog.mdx:409 — powerful ("the same powerful automation capabilities")
• changelog.mdx:453 — powerful ("The powerful GPT Image 1.5 model")
• changelog.mdx:455 — leverage ("Leverage OpenAI's latest model")
• changelog.mdx:481 — powerful ("troubleshoot errors with powerful visual analytics")
• changelog.mdx:513 — powerful ("Anthropic's most powerful AI capabilities")
• changelog.mdx:571 — powerful ("our powerful Evals feature")
• changelog.mdx:603 — powerful ("apply powerful content safety controls")
• changelog.mdx:641 — seamlessly ("Integrate seamlessly with major observability")
• changelog.mdx:821 — powerful AND effortlessly ("Slide Builder gets a powerful upgrade … allowing you to effortlessly rearrange slides")

🧩 Component suggestions (2)

• changelog.mdx:208,234,259,287,313,339,365,397,423,449,479,509,539,569,601,631,661,691,723,753,785,817 — 22 entries use **Bold Title Case Text** as the section heading for each feature instead of a proper ### heading. These should become ### Feature name in sentence case so they render as headings and are consistent with the May–July 2026 entries which already do this correctly.

• changelog.mdx:112 and throughout — 120 ➡️ emoji arrows are used in place of bullet points across all pre-May 2026 entries. CLAUDE.md prohibits emojis in documentation content. Replace with standard - list items.

⚠️ Contradictions (1)

• changelog.mdx:20 says "See the Evals documentation for full details" about Tool Argument Matching, but evals.mdx:105 only enumerates four check types: LLM Judge, Text Includes, Text Equals, and Tool Usage. Tool Argument Matching appears nowhere in the Evals docs. Either update evals.mdx to document the new check type before this changelog entry ships, or soften the pointer ("the Evals documentation will be updated shortly") so readers aren't left with a broken promise.

🔋 Credit usage

Item	Count
Files reviewed	1
Context pages read	2
Total lines processed	~1,010

Files read: changelog.mdx (854 lines), build/agents/build-your-agent/evals.mdx (~160 lines sampled), build/invent/invent.mdx (~90 lines sampled)

[mintlify]

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

Project	Status	Preview	Updated (UTC)
relevanceai	🟡 Building	–	Jul 15, 2026, 6:24 AM

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