Feature request: keep a non‑agent codebase index for manual diff workflow

[mynamemyway]

Feature request: bring back non‑agent codebase index for manual diff workflow

Hi,

I`m Alex a backend developer using VS Code and testing Continue as a replacement for Gemini Code Assist. I tested many analogs and i really like Continue: it's convenient, fast, and greatly implemented.

My primary workflow is very manual and non‑agentic:

• chat → agree on a plan → generate diffs → review → apply / discard per file
• the assistant needs to understand the whole repository, but I explicitly control when and how context is used, to avoid wasting tokens/requests or unnecessary actions that agents love to impose.

The old @Codebase index was conceptually a perfect fit:

• index the repo once
• then use fast semantic retrieval for chat/edits
• without an autonomous agent walking the filesystem on every request

With the new Agent mode + tools approach and deprecation of @Codebase / @Docs, there is no way to:

• keep a persistent codebase index that the chat mode can use
• do non‑agent, token‑efficient work on large repos
• have “manual” control of when the model reads which files (for example, when I pay per request or have strict daily limits)

Agent mode is great for some users, but for my use‑case:

• I do not want the agent to automatically traverse the repo, run tools, or change files
• I just need a stable, indexed view of the codebase that chat mode can query when I ask for diffs or refactors

Would it be possible to:

• keep (or re‑introduce) a non‑agent codebase index as an optional feature; or
• expose a way for chat mode to use a local code‑RAG index (embeddings + vector store) without enabling Agent mode / MCP tools?

This would make Continue extremely powerful and quite unique for users who:

• prefer manual, deterministic workflows over agents
• work with strict external quotas (e.g. Mistral, Gemini, Claude) and need to minimise unnecessary context / tool calls in order to keep everyday work efficient for both developers and model providers
• use the IDE mostly for precise backend work, with one diff per file per iteration

Right now I can use Qwen or other tools for full‑repo code review, and Continue only for small feature work with manually added files, but having a proper non‑agent index would make Continue my primary tool.

Thanks for considering this, and happy to share more details about my setup and usage pattern if helpful.