Repository Intelligence

Repository Intelligence searches local code and returns verbatim source lines for a question instead of generating prose; relevance is not guaranteed.

Note

Status: Experimental research prototype (v0.1.1) for local code inspection.

• Searches code using text matching, offline hash embeddings, or local neural models.
• Responds to questions by quoting exact source lines instead of generating text; quotes match the source, but relevance is not guaranteed.
• Stores portable repository index snapshots without external database dependencies.
• Synchronizes file additions, changes, and deletions incrementally through Git diffs.
• Exposes local search and index reload endpoints over a lightweight HTTP server.

Quick start

Try the released CLI (macOS or Linux)

The prebuilt CLI does not require Rust. The installer downloads a release, checks its SHA-256 checksum, and installs to ~/.local/bin:

curl -fsSL https://raw.githubusercontent.com/rustfuture/repository-intelligence/main/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"
repository-intelligence --version
# Search the repository you are in, offline, and print the three best source lines
repository-intelligence --hybrid . "main function" --top 3

Prebuilt Linux binaries require glibc; unsupported platforms can build from source. On Windows, use the PowerShell installer.

Build from source

You need Git and Rust 1.85+ with Cargo (rustup). It has no third-party crate dependencies. Offline commands use a deterministic HashEmbedding provider without external services. Neural embeddings and model-assisted answers use a local Ollama service (nomic-embed-text and qwen2.5-coder:1.5b) or the AGY adapter (the Google Antigravity command-line client).

git clone https://github.com/rustfuture/repository-intelligence.git
cd repository-intelligence

# Build binary
cargo build --locked

# View available CLI options and modes
cargo run --locked -- --help

# 1. Index a repository (default retrieval uses deterministic HashEmbedding)
cargo run --locked -- --index evaluation/corpus /tmp/ri.ri

# 2. Retrieve evidence offline without calling a model
cargo run --locked -- --semantic evaluation/corpus "index header"
cargo run --locked -- --hybrid   evaluation/corpus "index header"
cargo run --locked -- --hybrid evaluation/corpus "index header" --top 3

# 3. Query a saved snapshot
cargo run --locked -- --load-index /tmp/ri.ri "index header"

The offline examples print file and line references from the bundled corpus; the snapshot is written to /tmp/ri.ri. No API key, running model service, or model weights are needed for those examples. Run them from the repository root. These commands use macOS/Linux paths; choose a local snapshot path on Windows.

Optional model-assisted answers

Start Ollama and download qwen2.5-coder:1.5b before running these commands. See the Ollama quickstart for installation. This step needs model weights and is separate from the offline examples.

# Extractive answer: the model returns evidence IDs, the app renders the source
USE_OLLAMA=1 cargo run --locked -- --answer evaluation/corpus "What static string does reload return in api.rs?"

# Machine-readable answer record
USE_OLLAMA=1 cargo run --locked -- --answer-json evaluation/corpus "What static string does reload return in api.rs?"

Default retrieval uses deterministic HashEmbedding (not a learned model). RI_EMBEDDING_PROVIDER=nomic selects local neural embeddings. USE_OLLAMA=1 selects Ollama; without it the AGY adapter (Google Antigravity command-line client) is used. No automatic hash fallback occurs when a selected neural provider fails.

Architecture

flowchart TD
    R[Repository] --> S[Filtered Scanner]
    S --> C[Chunks / Spans]
    C --> L[Lexical Index]
    C --> V[Vector Index]
    L --> H[Hybrid Retrieval - RRF]
    V --> H
    H --> E[Evidence Candidates]
    E --> M[Model Selects Evidence IDs]
    M -->|IDs: E1, E2| O[Application Renders Original Source]
    M -->|Refusal| N[Return No Answer]

Loading

• A repository scanner filters binary and sensitive files, then chunks source files into line-addressed spans.
• Lexical matching and vector embeddings index chunks, and reciprocal rank fusion combines candidate scores.
• Top candidates are formatted into numbered evidence blocks ([E1], [E2]) and passed to the model prompt.
• The model selects only evidence identifiers and cannot return free-form prose.
• The application verifies selected identifiers and displays original source lines directly from disk.

For component details, trust boundaries, and system overview, see docs/architecture.md.

Answer Contract

The --answer and --answer-json commands use extractive-selection-v1. The model returns only evidence IDs. The application validates each ID and renders original source text with file and line references. Extra prose, invalid IDs, and mixed valid/invalid output are refused atomically.

An exact quotation is not proof that a source is true or answers the question. Selections may be irrelevant, and repository text is treated as untrusted data. Output is labelled as source excerpts, not verified answers. The model receives no shell or tool execution privileges.

Measured Results

Full v3 regression record: 42 previously exposed questions (30 answerable, 12 unanswerable; 41 model calls, 1 pre-model refusal). Results yielded 12 false accepts (5 irrelevant, 2 partial, 5 on unanswerable questions) and 2 false rejects, with 21 expected sources fully covered and 7 correct refusals.

Three runs of these 42 questions are committed (evaluation/v3/run-01, run-02, run-03). Run-03 is reported here because it is the latest run and the only one made after the unused src/citation.rs module was removed (commit 79fc4c1, clean tree); run-01 was made on a dirty working tree and run-02 on commit 63c1eba. The runs differ: false accepts / expected sources covered / false rejects are 12 / 18 / 3 (run-01), 8 / 23 / 3 (run-02) and 12 / 21 / 2 (run-03), so run-02 is the best of the three and run-03 is not the most favourable. Run-01 and run-02 record identical source, question and corpus hashes and model digests, so that spread is run-to-run variation rather than a recorded code change; treat these counts as one sample, not a stable rate.

The committed HTTP benchmark (evaluation/v3/benchmark-local-http.json) measures 200 sequential localhost requests on the fixed authored corpus at p50 0.857 ms / p95 0.920 ms (request overhead only, not generation latency).

All accepted excerpts matched source text (quotation integrity, not semantic correctness). For the extended results table and historical notes, see docs/measured-results.md.

Reproducibility

Reproduce the evaluation into a new directory (existing outputs are never overwritten):

Reproduction details

python3 evaluation/v3/evaluate.py --output evaluation/v3/my-run
python3 evaluation/v3/evaluate.py --render-only evaluation/v3/my-run   # re-render, no model calls

The output summary.json separates false_accepts, false_rejects, model_called, and pre_model_refusals. A pre-model refusal indicates the index lacked a lexical anchor and the model was not called. Historical notes on superseded claims are relocated to docs/measured-results.md.

Tests

cargo fmt --check
cargo check --locked --all-targets
cargo clippy --locked --all-targets -- -D warnings
cargo test --locked
python3 -m unittest discover -s evaluation/v3 -p 'test_*.py'
python3 evaluation/evaluate.py
python3 evaluation/evaluate_modes.py

More test commands

The test suite validates formatting, static analysis, 43 Rust unit and regression tests covering indexing and retrieval, and 5 Python evaluator tests.

Limitations

• Quotation integrity is not correctness. Matching source text proves the excerpt exists; it does not prove the excerpt answers the question.
• Small, authored, previously exposed corpus. The 42 v3 questions are regression evidence, not an unseen benchmark.
• Generic overlap scoring. Expected spans are frozen and overlap-based; entailment is not measured automatically.
• No isolation. The HTTP service has no TLS, auth, or tenant separation. Bind to loopback.
• A synthesis or NLI layer would need its own evaluation. A larger model or an NLI judge does not by itself guarantee correctness.

Design references

• ALCE (EMNLP 2023) — citation presence and citation support are different evaluation dimensions.
• Anthropic long-context experiments — quote extraction can make source use more inspectable.

Repository Map

Repository map

Path	Contents
src/lib.rs	Index, retrieval modes, git sync, evidence assembly
src/extractive.rs	extractive-selection-v1 validation and rendering
src/llm.rs	Ollama and AGY provider adapters
src/main.rs	CLI and local HTTP service
tests/ri_regression.rs	End-to-end regression tests
evaluation/corpus/	Authored Rust/Markdown corpus
evaluation/v3/	Evaluator, questions, and committed run records
docs/architecture.md	Component and trust boundaries

License

MIT — see LICENSE. The evaluation corpus is authored for this project.