docs(llm): clarify Vertex AI configuration options

[mason5052]

Summary

Clarify in the user-facing LLM provider docs that PentAGI does not currently expose a dedicated Google Vertex AI configuration path for Anthropic Claude in .env, and describe the routes that actually work today (direct Anthropic, AWS Bedrock, or a custom OpenAI-compatible gateway).

Docs-only change. No Go code, no installer behavior, no new environment variables.

Problem

Issue #310 asks how to provide a Google Vertex AI API key in .env for Anthropic Claude. PentAGI's .env.example, backend/pkg/config, and backend/cmd/installer/wizard do not read any VERTEX_*, GOOGLE_APPLICATION_CREDENTIALS, or vertex_ai variable. The current provider initialization for Claude only knows about:

• ANTHROPIC_API_KEY / ANTHROPIC_SERVER_URL (direct Anthropic API)
• BEDROCK_REGION / BEDROCK_DEFAULT_AUTH / BEDROCK_BEARER_TOKEN / BEDROCK_ACCESS_KEY_ID / BEDROCK_SECRET_ACCESS_KEY / BEDROCK_SESSION_TOKEN / BEDROCK_SERVER_URL (AWS Bedrock)
• LLM_SERVER_URL / LLM_SERVER_KEY / LLM_SERVER_MODEL (custom OpenAI-compatible endpoint)

Today this leaves users guessing whether some hidden Vertex AI configuration exists. The user in issue #310 hit exactly this — they "tried multiple methods of adding Anthropic's Claude models Vertex AI keys but the application couldn't parse it."

Solution

Add a small clarification in two places, matching the existing style of each file:

• README.md: a > [!NOTE] callout inside the Anthropic Provider Configuration section. It states the current limitation, lists the supported Claude routes (direct Anthropic, AWS Bedrock), and describes the safest workaround for users who must use Vertex AI today (front it with an OpenAI-compatible gateway and configure the existing Custom LLM provider, with a caveat that reliability depends on the gateway).
• backend/docs/config.md: a matching Note paragraph under the Anthropic section, pointing at the AWS Bedrock LLM Provider and Custom LLM Provider sections already in the same doc, and stating that no VERTEX_API_KEY or GOOGLE_APPLICATION_CREDENTIALS variable is wired into provider initialization today.

Wording is intentionally hedged ("currently", "at this time", "today") to avoid permanent claims, and no fake env vars are introduced.

User Impact

• Users following issue How to provide Google Vertex AI API key in the .env file #310 get a direct, accurate answer in the docs they were already reading.
• New users evaluating PentAGI for Claude-on-Vertex-AI workloads see the limitation up front and the supported workaround, instead of trial-and-erroring .env keys that do nothing.
• No runtime change: existing deployments keep working unchanged.

Test Plan

• git diff --check clean
• git status shows only README.md and backend/docs/config.md modified; no unrelated dirty files staged
• Verified every env var named in the new text exists in code or existing docs:
  • ANTHROPIC_API_KEY, ANTHROPIC_SERVER_URL: backend/pkg/config/config.go, .env.example, README, backend/docs/config.md
  • BEDROCK_REGION, BEDROCK_DEFAULT_AUTH, BEDROCK_BEARER_TOKEN, BEDROCK_ACCESS_KEY_ID, BEDROCK_SECRET_ACCESS_KEY, BEDROCK_SESSION_TOKEN, BEDROCK_SERVER_URL: backend/pkg/config/config.go, .env.example, README, backend/docs/config.md
  • LLM_SERVER_URL, LLM_SERVER_KEY, LLM_SERVER_MODEL: backend/pkg/config/config.go, .env.example, backend/docs/config.md
• Verified VERTEX_*, vertex_ai, and GOOGLE_APPLICATION_CREDENTIALS are NOT present in backend/pkg/config or backend/pkg/providers — the doc claim that no such variable is wired in today is accurate
• Anchor links resolve: #aws-bedrock-provider-configuration (README) and #aws-bedrock-llm-provider + #custom-llm-provider (config.md) match real headings in the same files
• No runtime Go files, installer behavior, or generated files modified

Refs #310

[Copilot]

Pull request overview

Updates PentAGI’s LLM provider documentation to explicitly clarify that Anthropic Claude via Google Vertex AI is not currently configurable through dedicated .env variables, and to point users to the supported configuration paths that work today (direct Anthropic, AWS Bedrock, or a Custom LLM gateway workaround). This addresses confusion raised in issue #310 without changing runtime behavior.

Changes:

• Add a GitHub NOTE callout in README.md under Anthropic provider configuration explaining the lack of Vertex AI .env wiring and listing supported Claude routes.
• Add a matching “Note on Google Vertex AI” in backend/docs/config.md, including links to the AWS Bedrock and Custom LLM provider sections and explicitly stating no VERTEX_API_KEY / GOOGLE_APPLICATION_CREDENTIALS wiring exists today.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
README.md	Adds an Anthropic section NOTE clarifying Vertex AI isn’t directly configurable and describes supported alternatives/workaround.
backend/docs/config.md	Adds an Anthropic note clarifying no Vertex AI env vars are wired and points to Bedrock/Custom LLM provider configuration sections.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[asdek]

thank you for the PR!