AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.
Rust agent-memoryai-agentsbm25claude-codeclicode-searchcursordatabaseelasticsearchfull-text-searchhybrid-searchjsonllocal-firstmcpmcp-servermodel-context-protocolragrustsearch-enginevector-search
Rust AST security substrate for XERJ
Tools that turn a Rust workspace into a queryable security substrate in XERJ, so
an AI agent can interrogate the code instead of reading all of it. Built to
audit XERJ's own engine — see the case study in
../../case-studies/xerj-self-audit/.
Scripts
rust_ast_index.py— tree-sitter-rust extractor. Walks a workspace and emits three NDJSON streams (functions,calls,routes) pluscoverage.jsonandunsafe_inventory.json. Records carry the security-relevant AST facts: unsafe ops, panic/abort sites,ascasts, allocation-from-parameter shapes, axum extractors, filesystem/process/deserialize sinks, validator calls, and lock-across-await. 100% function coverage on XERJ's engine, asserted against an independent AST node count.ingest.py— creates three XERJ indices with explicit mappings (exact values arekeyword, code bodies aretext) and bulk-loads the streams, asserting sent == indexed.find_recursion_cycles.py— the query grep cannot express: strongly- connected components of the free-function call graph, flagging cycles with no depth guard (the stack-overflow-from-nesting shape). Method-call edges are excluded on purpose (unresolvable without type inference).
Quick start
python3 -m venv venv && . venv/bin/activate
pip install tree-sitter==0.26.0 tree-sitter-rust==0.24.2
python3 rust_ast_index.py ../../../engine --out ./ast-out
# boot xerj on 9310 (see the case study REPRODUCE.md), then:
python3 ingest.py ./ast-out --url http://127.0.0.1:9310
python3 find_recursion_cycles.py ./ast-out
Audit queries with live hit counts: QUERY_COOKBOOK.md.
Committed sample outputs: sample-coverage.json, sample-unsafe-inventory.json.