Question 1

Why not just use ripgrep?

Accepted Answer

Ripgrep is exceptionally fast at finding text patterns, but it has no semantic understanding of code. It doesn't know the difference between a variable declaration and a string literal, nor can it follow call trees. CostAffective uses Tree-sitter AST queries to build structural maps, returning precise code scopes and relationships instead of raw, noisy regex matches.

Question 2

Why not use CodeGraph?

Accepted Answer

CodeGraph builds heavy, memory-resident dependency graphs that can bloat context sizes and consume massive RAM. CostAffective uses a lightweight SQLite index and returns compressed, symbol-level AST contexts, lowering daily prompt token usage by 45.9% compared to traditional code graphs.

Question 3

What is repository intelligence?

Accepted Answer

Repository intelligence is the process of statically parsing and indexing codebases to map structure (definitions, types, call graphs, references) into a querying system. This allows AI editors and coding agents to navigate code bases dynamically and accurately, without reading every file.

Question 4

How much context can CostAffective save?

Accepted Answer

In global leaderboard evaluations, CostAffective reduced token consumption by 45.9%, exploration loops by 54.3%, and tool interaction latency by 88% compared to standard naive context retrievals.

Question 5

Does it work with Codex CLI?

Accepted Answer

Yes. CostAffective is fully compatible with Codex CLI, Claude Code, Cursor, OpenCode, Antigravity, and any other IDE client supporting the standard Model Context Protocol (MCP).

Question 6

Does my code leave my machine?

Accepted Answer

No. CostAffective is 100% local-first. All AST parsing, index updates, search queries, and database lookups run locally on your system. No network egress, no telemetry, and no external code leaks.

Question 7

Do I need an LLM API key to index repositories?

Accepted Answer

No. Indexing is performed statically using local tree-sitter AST parsers and SQLite compilation. No LLM APIs or cloud tokens are consumed during codebase indexing.

Question 8

How does the watchdog update indexes?

Accepted Answer

CostAffective integrates a system directory watchdog (using fsnotify). When a file save occurs, the watchdog computes the file's hash delta and updates only the modified files in the SQLite database, completing incremental updates in under 8ms.

Question 9

What is the SQLite database file footprint?

Accepted Answer

The local SQLite index is highly optimized. On average codebases (e.g., 500 files), the database file size is less than 20MB, and memory footprint during search queries is under 15MB.

Question 10

What platforms are supported? Does it run on macOS, Windows, and Linux?

Accepted Answer

Yes. CostAffective binaries are compiled for Linux (amd64) and Windows (amd64). macOS and Linux ARM architectures can be built from source using local Go compilers.

Question 11

What languages does CostAffective support?

Accepted Answer

CostAffective currently parses Go, Python, TypeScript, JavaScript, Rust, C++, C#, Java, and HTML/CSS out-of-the-box using official Tree-sitter parsers.

Question 12

What MCP tools does CostAffective expose?

Accepted Answer

It exposes six core tools: search_code, find_symbol, find_references, find_callers, grep_code, and get_repository_summary.

Question 13

How does search_code differ from grep_code?

Accepted Answer

`search_code` leverages tree-sitter AST maps to identify functional blocks matching semantic search intents. `grep_code` is a full-text regex scanner used as a fallback for strings, comments, and untracked code chunks.

Question 14

Can it index large monorepos?

Accepted Answer

Yes. CostAffective is designed to handle monorepos. It leverages fast SQLite transactions and selective indexing, easily scaling to codebases containing thousands of source files.

Question 15

How do I configure CostAffective for Cursor?

Accepted Answer

Add CostAffective to Cursor's settings under "MCP" by specifying command "costaffective" and argument "serve" with stdio transport.

Question 16

Does it require Docker?

Accepted Answer

No. CostAffective runs as a standalone compiled native binary. It has zero dependencies on Docker, Docker containers, or node runtimes.

Question 17

What does the costaffective doctor command validate?

Accepted Answer

`costaffective doctor` checks binary permissions, PATH environment variables, MCP configuration maps in user directories, SQLite read/write states, and active server run status.

Question 18

Does it consume CPU cycles when idle?

Accepted Answer

No. The background watcher relies on passive OS file notification triggers. CPU cycles are consumed only for brief moments (usually <10ms) when files are modified.

Question 19

Can I customize the token budget?

Accepted Answer

Yes. AI clients can supply token budgets as parameters to MCP tools, forcing CostAffective to compress retrieval chunks if they exceed context limits.

Question 20

How do I add CostAffective to Claude Code?

Accepted Answer

CostAffective integrates automatically. You can install it by configuring settings in `~/.claude.json` or by running the auto-setup command `costaffective install --target claude`.

Question 21

Does it index third-party vendor dependencies?

Accepted Answer

By default, CostAffective ignores vendored folders (like node_modules or vendor/) to keep the index clean, but you can configure files to track internal libraries.

Question 22

Is CostAffective free and open-source?

Accepted Answer

Yes. CostAffective is released under the permissive MIT license and its repository is fully hosted on GitHub.

Question 23

Does it support custom ignore globs?

Accepted Answer

Yes. You can configure directories, file extensions, and size thresholds in the configuration files to skip indexing unwanted files.

Question 24

What is static alias resolution?

Accepted Answer

It is the technique of mapping import paths and structural aliases statically, allowing the resolver to find correct declarations even if import paths use directory alias pointers (e.g. `@/components`).

Question 25

What is the Model Context Protocol (MCP)?

Accepted Answer

The Model Context Protocol (MCP) is an open standard created by Anthropic that connects AI models to secure local or remote data sources and developer tools.

Question 26

Are search caches isolated across sessions?

Accepted Answer

Yes. The session manager isolates query caches, temporary index modifications, and search history boundaries for every active developer workspace.

Question 27

How do I build CostAffective from source?

Accepted Answer

Clone the repository and run `go build ./cmd/costaffective`. This compiles a standalone binary in the current directory.

Question 28

What happens if files change during a chat session?

Accepted Answer

The incremental watchdog registers the changes and updates the index. The next tool call issued by the coding agent instantly queries the fresh code state.

Question 29

Can I export the database index?

Accepted Answer

Yes, the index is stored as a standard SQLite `.db` file in the repository cache, allowing users to copy it to speed up first-time indexing on other machines.

Question 30

What is "Save tokens. Buy Coffee"?

Accepted Answer

It is a reminder that prompt tokens are expensive. By saving 45.9% on prompt contexts, developers can convert their cloud spending into real coffee.

Frequently Asked Questions