Axis search tools are the primary way agents avoid reinventing patterns and duplicating work. The agent protocol requires reaching for search by default before creating files or refactoring — not because the user asked, but because understanding what already exists produces better code. Local mode answers instantly from ripgrep with no index required. Hosted Pro mode blends in vector, full-text, and trigram retrieval fused and LLM-reranked, enriched with related files and symbol definitions.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/virsanghavi/axis/llms.txt
Use this file to discover all available pages before exploring further.
Free tier users get
search_codebase with local ripgrep — no index needed, instant results. Pro tier blends hosted vector results in when they arrive within budget. An upgrade prompt appears when a free org reaches for hosted search tools.search_codebase
Hybrid search over the indexed codebase. In local mode, the tool answers immediately from ripgrep plus keyword ranking — no index required. In hosted Pro mode, it blends vector + full-text + trigram retrieval, fuses the result sets, and applies LLM reranking. Pro results are additionally enriched with related files and symbol definitions.
A natural language question or code search query. Both “where is X” lookups and conceptual “how does Y work” queries are accepted, though
deep_search is the better tool for multi-file conceptual questions.The project to search. Defaults to the auto-detected project from the active workspace.
| Field | Description |
|---|---|
hits | Ranked results, each with file, line, and snippet. |
related | Files that historically change together with the top hits (hosted only). |
definitions | Symbols that the top hits call into — useful for tracing a call chain (hosted only). |
search_codebase for direct lookups; use deep_search for multi-file reasoning questions.
deep_search
Hosted Pro only.
deep_search is not available on the local stdio server. Local users get search_codebase with instant ripgrep-based results.deep_search reasons across files over multiple steps and returns a single cited answer. Every claim in the answer is grounded with a file:line reference.
A natural language question suited for multi-file reasoning. Examples: “How does token refresh work end to end?”, “Where is rate limiting enforced and why is it placed there?”, “What happens when a job claim fails?”
The project to search. Defaults to the auto-detected project.
file:line citations for every factual claim. The answer is produced after the engine has read across the relevant files — expect it to be slower than search_codebase but far more complete for architectural questions.
When to use: “How does X work?”, “Why is Y done this way?”, “What is the full flow for Z?” Use search_codebase when you need a quick “where is X” lookup — it is faster. Use deep_search when you need to understand something that spans multiple files.
index_codebase
Builds or refreshes the searchable index for the project. The indexer is incremental and content-hashed — unchanged files are compared by hash server-side and skipped, so re-running index_codebase is cheap even on large repos.
An array of
{path: string, content: string} objects to index. If omitted, the local server reads files from disk. Hosted callers pass file content explicitly because the hosted server has no access to the client’s filesystem.When
true, removes index entries for files that no longer exist. Use together with allPaths to identify deletions accurately.The complete list of current file paths in the repository. Used alongside
prune: true — any indexed path not present in allPaths is removed from the index.The project to index. Defaults to the auto-detected project.
- Once to build the initial index for a new project
- After large refactors or branch switches that change many files
- Agents call
index_file(see below) after individual file writes to keep the index current mid-session
For the initial index of a large repository, the
axis index CLI is faster than calling index_codebase from an agent. It walks the repo respecting .gitignore and uploads only content-changed files as deltas. Run npx axis-init@latest and then axis index from the repo root.index_file
Local only.
index_file reads file content from the client’s disk. On the hosted surface, push updated content through index_codebase’s files parameter instead.The path of the file to index.
The file content to index. If omitted, the local server reads the file from disk at
filePath.The project to index the file under. Defaults to the auto-detected project.
search_docs
Local only.
search_docs is backed by the local RAG fallback over indexed documentation files. A hosted docs search is not yet available.A natural language documentation search query.
The project whose documentation index to search.
Search workflow
Build the initial index
Before agents can use hosted search, the codebase needs to be indexed once. The fastest path for a new project is the CLI:Alternatively, ask any connected agent to call
index_codebase — it will walk the repo and upload only content-changed files.Search before writing
Before creating a new file or refactoring an existing one, run a search:Check the
related files in the result — files that historically change together are often files that need to be updated together.Use deep_search for architectural questions
When you need to understand a flow end to end, prefer Every claim in the returned answer carries a
deep_search:file:line citation you can verify.Local vs. hosted capabilities
Local (free)
search_codebase: ripgrep + keyword ranking- No index needed — instant results
index_codebaseandindex_filefor local indexingsearch_docsfor local documentation- No
deep_search
Hosted Pro
search_codebase: vector + full-text + trigram, fused and LLM-rerankedrelatedfiles anddefinitionsenrichmentdeep_search: multi-hop agentic answer engine with citations- Incremental indexed uploads (content-hashed, cheap re-runs)