docs(9ae0a626-0e4d-4950-9cb8-bf0e945d2750_8eba1583-5a00-476e-adfa-fa31e4f03716): document BYOK zero-credit behavior in Evals

[claude]

Summary

Adds a <Note> callout to the Cost and billing section of build/agents/evals/introduction.mdx explaining that components using BYOK (Bring Your Own Key) models show 0 credits in the eval cost breakdown.

The callout is placed after the component breakdown table and covers all three components:

• Scenario Runner model
• Agent or Workforce's own model
• LLM judge model used in Checks

Verification

This feature is live and reachable — users can configure BYOK models via project keys for agents today, and select BYOK models for LLM judge Checks.

Facts are verified from apps/nodeapi/src/llm/model_handler_helpers.ts lines 92–96, where CalculateModelCredits returns credits_cost: 0 and ubb_credit_cost: 0 when user_key_used is true.

Linear issue

https://linear.app/relevance/issue/9ae0a626-0e4d-4950-9cb8-bf0e945d2750_8eba1583-5a00-476e-adfa-fa31e4f03716/

[mintlify]

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

Project	Status	Preview	Updated
relevanceai	🟢 Ready	View Preview	Sep 16, 2026, 2:14 AM

💡 Tip: Enable Automations 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, correct capitalization throughout, American spelling. The <Note> on line 90–92 is three sentences — CLAUDE.md says callouts must be "a single short paragraph."
🟢	Technical clarity	9/10	Good gotchas (usage-based vs legacy plan difference on line 75, BYOK zero-credit behavior on line 90). Minor: missing blank line before --- on line 62 may render incorrectly.
🟢	Non-technical clarity	9/10	The "Why use Evals?" section (line 33–37) is clear and motivation-first. The non-determinism explanation is approachable.
🟡	Structure	8/10	CardGroup (lines 39–49) and Overview bullet list (lines 53–62) overlap — both cover Tests, Checks, and Monitor — but the CardGroup omits Runs and Publish.

Score key: 🟢 9–10, 🟡 6–8, 🔴 1–5.

✨ Overall vibe: Solid introduction page — clear definition, good motivation section, and a thorough cost/billing breakdown that will genuinely help users. Two small structural issues: the CardGroup and Overview section double-cover the same three features without covering the other two, and the <Note> callout is a bit long for its container.

🔧 Issues (2)

• build/agents/evals/introduction.mdx:62 — Missing blank line before ---. The - **Monitor** — ... bullet item and the horizontal rule are adjacent with no blank line between them. In CommonMark, --- following a paragraph without a blank line can render as a Setext H2 heading rather than a rule. Add a blank line after the list.

• build/agents/evals/introduction.mdx:90–92 — The <Note> callout is three sentences (BYOK explanation, reason, scope). CLAUDE.md specifies "a single short paragraph." Consider trimming to the essential point — e.g. "Components using a BYOK model show 0 credits because those calls draw from your own API key, not from Relevance AI-provided credits." — and folding the "This applies to..." scope note into the surrounding table or prose.

🏗️ Page structure (1)

• build/agents/evals/introduction.mdx:39–62 — The CardGroup (lines 39–49) covers three of the five Evals features: Test, Checks, and Monitor. The "Overview of Evals" bullet list (lines 53–62) then covers all five. This means the first three features get described twice in quick succession while Runs and Publish are only described once. If the CardGroup is meant as a visual teaser, either extend it to cover all five features, or remove it and let the Overview bullet list carry the full feature list on its own. As-is, readers hit the same content in two different forms before the page moves on.

⚠️ Contradictions (0)

No contradictions found between introduction.mdx and the context pages (test-sets.mdx, checks.mdx). Check limits (10 per Test), billing model, and feature descriptions are consistent across all three files.

🔋 Credit usage

Item	Count
Files reviewed	1
Context pages read	2
Total lines processed	~364

Files read: build/agents/evals/introduction.mdx (109 lines), build/agents/evals/test-sets.mdx (132 lines), build/agents/evals/checks.mdx (123 lines)