Skip to content

docs: align model references with actual agents (Opus 4.7)#511

Merged
jahooma merged 3 commits intoCodebuffAI:mainfrom
bogdan0083:docs/update-opus-4.7
Apr 20, 2026
Merged

docs: align model references with actual agents (Opus 4.7)#511
jahooma merged 3 commits intoCodebuffAI:mainfrom
bogdan0083:docs/update-opus-4.7

Conversation

@bogdan0083
Copy link
Copy Markdown
Contributor

Hey guys!

The docs were still telling users that Codebuff runs on Claude Opus 4.6 and referenced a few subagent models/modes that no longer exist. This PR brings the docs back in sync with what the agents in agents/ actually use today.

What changed

Opus 4.6 → Opus 4.7 across all user-facing docs:

  • tips/modes.mdx — mode table and Default/Max section text
  • advanced/what-models.mdx — orchestrator table and subagent model list
  • advanced/claude-code-comparison.mdx — orchestrator mention
  • advanced/how-does-it-work.mdx — orchestrator mention, plus Thinker/Editor/Reviewer rows corrected to match the current agent definitions (Thinker: Opus 4.7 + GPT-5.4, Editor: Opus 4.7 + GPT-5.1 + GLM 5.1, Reviewer: Opus 4.7 + GLM 5.1 in Lite)

FAQ answer rewrite (help/faq.mdx and the mirrored SEO structured data in web/src/app/docs/[category]/[slug]/page.tsx):

  • "Free mode" → "Lite mode" (Free was changed to Lite as I can see)
  • Drop stale "GPT-5.1 for deep reasoning", "Grok 4.1 Fast for terminal commands and research", and "Relace AI for fast file rewrites" mentions — file writes are handled deterministically by the runtime (packages/agent-runtime/src/process-file-block.ts), and Relace is only used as a file-picker re-ranker in admin tooling, not for edits
  • Replace with the current setup: Opus 4.7 + GPT-5.4 for reasoning/review, Gemini 3.1 Flash Lite for terminal commands, file discovery, and web/docs research

No behavior changes — docs only.

If some sentences don't sound right - please tell me, I'll update them.

Thanks! ❤️

Align docs with the models actually wired up in the agents:
- Opus 4.6 -> Opus 4.7 across modes, what-models, faq,
  how-does-it-work, and claude-code-comparison pages
- Correct Thinker/Editor/Reviewer model lists in
  how-does-it-work.mdx to match the current agent definitions
- Replace stale GPT-5.1 / Grok 4.1 / Relace mentions in faq.mdx
  with the Opus 4.7, GPT-5.4, and Gemini 3.1 Flash Lite setup
  currently in use

Made-with: Cursor
- Rewrite the "What model does Codebuff use?" answer to only
  list modes that actually ship (Default, Max, Lite) and the
  models actually in use (Opus 4.7, GPT-5.4, Gemini 3.1 Flash
  Lite), dropping stale mentions of Free mode and Relace for
  file rewrites (file writes are handled deterministically by
  the runtime).
- Update the mirrored FAQ entry in the docs page SEO structured
  data so it matches faq.mdx.

Made-with: Cursor
@greptile-apps
Copy link
Copy Markdown
Contributor

greptile-apps bot commented Apr 19, 2026

Greptile Summary

This PR updates user-facing documentation to align model references with what the agents in agents/ actually use, promoting Claude Opus 4.6 to Opus 4.7 across six doc files and renaming the "Free" mode to "Lite" in the FAQ and modes table.

Changes made:

  • tips/modes.mdx, advanced/what-models.mdx, advanced/claude-code-comparison.mdx, advanced/how-does-it-work.mdx — Opus 4.6 → Opus 4.7 throughout
  • help/faq.mdx — model answer rewritten to current stack; Claude Pro/Max subscription entry updated with deprecation warning
  • web/src/app/docs/[category]/[slug]/page.tsx — FAQ JSON-LD model answer updated

Issues found:

  • what-models.mdx lines 22 and 40 still use the old "Free" mode label instead of "Lite", inconsistent with every other file in this PR
  • how-does-it-work.mdx line 25 still lists the Researcher agent as running on Grok 4 Fast for web/docs lookup, while what-models.mdx and faq.mdx (both updated here) attribute that task to Gemini 3.1 Flash Lite
  • page.tsx lines 39-42 — the SEO JSON-LD answer for the Claude Pro/Max subscription question still promotes connecting a Claude subscription, directly contradicting the deprecation warning that faq.mdx now shows to users

Confidence Score: 3/5

Three inconsistencies remain: two 'Free' → 'Lite' renames were missed, the Researcher model is in conflict across files, and SEO JSON-LD still promotes a deprecated feature that the MDX now warns against.

The core Opus 4.6 → 4.7 updates are correct and consistent, but three concrete gaps remain. The most impactful is the page.tsx FAQ JSON-LD still advertising the Claude Pro/Max subscription integration as available while faq.mdx now actively discourages it — this is user-facing misinformation surfaced by search engines. The other two (Free → Lite rename and Grok vs Gemini for Researcher) are documentation inconsistencies that would confuse readers comparing sections.

web/src/content/advanced/what-models.mdx (Free → Lite), web/src/content/advanced/how-does-it-work.mdx (Researcher model), and web/src/app/docs/[category]/[slug]/page.tsx (stale Claude Pro/Max subscription JSON-LD answer)

Important Files Changed

Filename Overview
web/src/content/advanced/what-models.mdx Orchestrator table and subagent list updated from Opus 4.6 to Opus 4.7, but two "Free" mode references were not renamed to "Lite" (lines 22 and 40)
web/src/content/advanced/how-does-it-work.mdx Thinker/Editor/Reviewer model rows correctly updated, but Researcher agent still lists "Grok 4 Fast" while other updated docs attribute web/docs research to Gemini 3.1 Flash Lite
web/src/app/docs/[category]/[slug]/page.tsx FAQ model answer updated correctly, but the Claude Pro/Max subscription answer still promotes the feature that faq.mdx now marks as deprecated, causing SEO JSON-LD to contradict the visible page
web/src/content/tips/modes.mdx Model references cleanly updated from Opus 4.6 to Opus 4.7 across the mode table and Default/Max section headers; no issues found
web/src/content/advanced/claude-code-comparison.mdx Single orchestrator reference updated from Opus 4.6 to Opus 4.7; no issues found
web/src/content/help/faq.mdx Model answer rewritten to reflect current Opus 4.7 / Gemini 3.1 Flash Lite setup; Claude Pro/Max subscription entry updated with correct deprecation warning

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    U[User Prompt] --> ORCH

    subgraph Orchestrator
        ORCH["Buffy\nDefault/Plan/Max → Claude Opus 4.7\nLite → GLM 5.1"]
    end

    ORCH --> FP["File Picker\nGemini 2.0 Flash"]
    ORCH --> CS["Code Searcher\ngrep-style"]
    ORCH --> RES["Researcher\nGrok 4 Fast ⚠️ / Gemini 3.1 Flash Lite?"]
    ORCH --> TH["Thinker\nOpus 4.7 + GPT-5.4"]
    ORCH --> ED["Editor\nOpus 4.7 + GPT-5.1 + GLM 5.1"]
    ORCH --> REV["Reviewer\nOpus 4.7 (GLM 5.1 in Lite)"]
    ORCH --> BA["Basher\nGemini 3.1 Flash Lite"]

    ED --> SEL["Best-of-N Selector\n(Max mode only)"]
    SEL --> OUT[Final Output]
    REV --> OUT
Loading

Comments Outside Diff (2)

  1. web/src/content/advanced/what-models.mdx, line 40 (link)

    P1 Second "Free mode" reference missed

    The closing sentence also uses "Free mode" instead of "Lite mode", which is inconsistent with the rename done elsewhere in this PR.

  2. web/src/app/docs/[category]/[slug]/page.tsx, line 39-42 (link)

    P1 SEO structured data contradicts the updated FAQ content

    The FAQ_ITEMS entry for "Can I use my Claude Pro or Max subscription with Codebuff?" still contains the old promotional answer ("Yes! If you have a Claude Pro or Max subscription, you can connect it…"), while faq.mdx now says this feature is deprecated and actively warns users away from it (risk of account disablement, removal on March 1st).

    Search engines will index this JSON-LD data and show users the outdated, encouraging answer — the opposite of what the updated MDX communicates. This answer should mirror the deprecation warning in faq.mdx.

Reviews (1): Last reviewed commit: "docs: tighten FAQ model answer and fix S..." | Re-trigger Greptile

Comment thread web/src/content/advanced/what-models.mdx Outdated
Comment thread web/src/content/advanced/how-does-it-work.mdx Outdated
- what-models.mdx: rename "Free" -> "Lite" in the orchestrator
  table and the trailing paragraph so every doc in this PR uses
  the same user-facing mode label.
- how-does-it-work.mdx: the Researcher subagent is Gemini 3.1
  Flash Lite (agents/researcher/researcher-web.ts and
  researcher-docs.ts), not Grok 4 Fast. Matches what-models.mdx
  and faq.mdx.
- page.tsx: update the FAQ JSON-LD answer for the Claude Pro /
  Max subscription question to reflect the deprecation that
  faq.mdx already surfaces, so SEO structured data and the
  rendered page agree.

Made-with: Cursor
@jahooma
Copy link
Copy Markdown
Contributor

jahooma commented Apr 20, 2026

Thank you @bogdan0083! Looks good, will just merge it

@jahooma jahooma merged commit f5bbd99 into CodebuffAI:main Apr 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants