`java-codebase-rag` CLI — operator guide

The java-codebase-rag command is the operator surface for this bundle: index lifecycle (init / increment / reprocess / erase), graph and Lance health (meta, tables), ignore diagnostics, and PR diff analysis. It is not the MCP navigation surface (that is search / find / describe / neighbors / resolve on the MCP server — this CLI is lifecycle and introspection only). For agents driving the MCP server, see AGENT-GUIDE.md.

Install and discovery

After installing the package (e.g. editable install from the repo root), the console script is on your PATH:

.venv/bin/pip install -e .
java-codebase-rag --help

If java-codebase-rag is missing, run the module entrypoint:

.venv/bin/python -m java_codebase_rag.cli --help

Setup commands

`install`

Interactive setup wizard that walks users through Java source detection, embedding model selection, agent host configuration, artifact deployment, and YAML config generation. Use --non-interactive for CI/automation.

# Interactive mode
java-codebase-rag install

# Non-interactive mode (requires at least one --agent)
java-codebase-rag install --non-interactive --agent claude-code
java-codebase-rag install --non-interactive --agent claude-code --agent qwen-code

# With custom embedding model
java-codebase-rag install --model /path/to/model

# User-scope installation (available globally)
java-codebase-rag install --scope user

Flags:

--non-interactive — Run without prompts (requires --agent).
--agent {claude-code,qwen-code,gigacode} — Agent host to configure (can be passed multiple times).
--scope {project,user} — Installation scope (default: project). Project scope writes to .<host>/ in the project repo; user scope writes to ~/.<host>/ (globally available).
--model MODEL — Embedding model path or auto (default: auto, downloads sentence-transformers/all-MiniLM-L6-v2 on first run).
--quiet / -q — Suppress the indexing progress stream on stderr (wizard prompts unchanged).
--verbose / -v — Raw-relay subprocess output during the indexing sub-step (no progress bar).

Exit codes:

0 — Success (all stages completed).
1 — Partial success (some stages failed). Re-run install to retry failed stages.
2 — Fatal error (no Java files found, required flag missing).

Stages:

Java source detection — Maven/Gradle module roots.
Embedding model selection — auto-download or local path.
Agent host selection — Claude Code, Qwen Code, GigaCode (multi-select).
Install scope — project or user.
MCP entrypoint resolution + artifact deployment — config, skill, agent files.
Index + finish — YAML generation, .gitignore update, init. Stage 6's indexing sub-step renders the unified Vectors → Optimize → Graph progress on stderr (see Indexing progress); the wizard's conversational stdout is unchanged.

Re-running install: If .java-codebase-rag.yml exists, the installer shows current values and offers "Update" (pre-filled) or "Start fresh". Existing MCP entries are updated in-place (merged, not duplicated). Skill/agent files trigger overwrite confirmation.

`update`

Post-upgrade refresh: overwrites skill and agent files with the latest shipped versions and updates the MCP command path. If an index exists, also runs an incremental Lance + graph catch-up (same as increment). Requires a prior install run.

# Refresh after pip upgrade
pip install --upgrade java-codebase-rag
java-codebase-rag update

# Preview changes without writing
java-codebase-rag update --dry-run

# Force overwrite all artifacts
java-codebase-rag update --force

Flags:

--force — Overwrite all artifacts even if content matches.
--dry-run — Print changes without writing files.
--quiet / -q — Suppress the indexing progress stream on stderr (wizard stdout unchanged).
--verbose / -v — Raw-relay subprocess output during the indexing sub-step (no progress bar).

Behavior:

Detects previously configured agent hosts (scans both project-level and user-level config files).
Refreshes skill and agent files (versioned assets from the package).
Updates MCP entrypoint path if java-codebase-rag-mcp has moved.
Runs an incremental index update (Lance + graph) if an index exists — same as java-codebase-rag increment. The indexing sub-step renders the unified Vectors → Optimize → Graph progress on stderr (see Indexing progress); it no longer runs silently.

Exit codes:

0 — Success.
1 — Partial failure (some artifacts failed to write).
2 — No configured hosts found.

Output mode

TTY: human-readable pprint of the payload on stdout (except successful selective reprocess with --vectors-only / --graph-only, which prints Rebuilt: / Skipped: lines instead of dumping the full dict).
Piped / non-TTY: single JSON object per invocation on stdout (no trailing noise). Use this in scripts and CI.
Lifecycle stderr: init, increment, reprocess, install, update, and erase stream subprocess progress (and relayed child stdout) to stderr; pass --quiet to suppress that stream. stdout stays the JSON/pprint payload (init/increment/reprocess) or the wizard conversational text (install/update) only.

Example:

java-codebase-rag meta --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag | jq .ontology_version

Indexing progress (stderr)

All five lifecycle commands that build the index (init, increment, reprocess, install, update) render the same unified progress on stderr during indexing: a header line, a three-phase list Vectors → Optimize → Graph, and a footer line. The phase list is the single source of truth for "what's happening right now":

Vectors — the cocoindex update Lance catch-up / full reprocess.
Optimize — the serialized Lance table compaction that runs after a successful vectors phase.
Graph — the build_ast_graph.py LadybugDB/LadybugDB build (full or incremental).

Determinate vs indeterminate per command:

Phase	Determinate?
`Vectors` (full `init` / `reprocess`)	Approximately determinate — a pre-walk estimates the file count; the bar clamps to 100% on completion (the pre-walk overstates by ignored/empty files).
`Vectors` (incremental `increment` / `update`)	Indeterminate — CocoIndex's `memo=True` cache only calls the per-file function for changed files, so no denominator is known up front. A pulsing bar plus a "files touched: N" counter.
`Optimize`	Always indeterminate (no item count exposed by Lance compaction).
`Graph` (full `init` / `reprocess`)	Determinate — pass 1 does a count-first filtered walk for an exact total; passes 2–6 are six known steps.
`Graph` (incremental `increment` / `update`)	Determinate when it runs; falls back to a full rebuild on schema change.

Flags, TTY, and failure:

Mode	Behaviour
TTY (default)	`rich` `Live` region — the multi-line phase display (spinner + bar + `%` + ETA).
Non-TTY / CI	`rich` auto-disables; concise throttled stderr lines (~every 5 s per phase + a terminal line) so CI logs still show progress.
`--quiet` / `-q`	Suppresses the entire progress stream (no header, phases, or footer). The stdout payload is unchanged.
`--verbose` / `-v`	Bypasses parsing; relays raw subprocess output verbatim (Lance warnings, brownfield events, the raw `JCIRAG_PROGRESS` protocol lines). No `Live` region.
Phase failure	The failing phase renders a red `✗`; the footer carries `(exit=N)`. The `rich` `Live` region is torn down cleanly so the error stays visible.
Missing `cocoindex` / builder binary	The pre-spawn stub emits a `status=failed` line; no phase is left hung at `running`.

Behaviour change (this release). install and update now emit their indexing progress on stderr (previously install printed indexing chatter to stdout, and update ran the whole indexing step with quiet=True — completely silent). The wizard conversational stdout for both commands is otherwise unchanged. update's previously-ignored --quiet / --verbose flags, and install's previously-ignored --verbose flag, are now wired through (install already honored --quiet).

Environment variables (summary)

Variable	Role
`JAVA_CODEBASE_RAG_INDEX_DIR`	Root directory for Lance tables, the LadybugDB file `code_graph.lbug`, and default cocoindex state. Default: `./.java-codebase-rag/` under the resolved Java tree root. Overridden by `--index-dir` or YAML `index_dir:`.
`SBERT_MODEL` / `SBERT_DEVICE`	Embedding model and device; must match the index. Overridden by `--embedding-model` / `--embedding-device` or YAML `embedding.model` / `embedding.device`.
`JAVA_CODEBASE_RAG_DEBUG_CONTEXT`	Verbose stderr logging for context expansion (diagnostic).
`JAVA_CODEBASE_RAG_RUN_HEAVY`	Test-only gate for slow end-to-end indexer tests (`pytest`).

Precedence (when a knob exists in more than one place): CLI flag > env var > YAML (.java-codebase-rag.yml) > built-in default.

Only the variable names in the table above are read as configuration.

Shared flags

Every subcommand accepts (all optional unless noted):

Flag	Meaning
`--source-root DIR`	Java repository root (default: current working directory).
`--index-dir DIR`	Index directory (default: `./.java-codebase-rag` under the resolved source root, or `JAVA_CODEBASE_RAG_INDEX_DIR`).
`--embedding-model` / `--embedding-device`	Override embedding resolution for subprocesses that honor env.

LadybugDB always resolves to <index-dir>/code_graph.lbug.

Relative paths for diagnose-ignore <path> are resolved against the MCP/CLI project root helper (--source-root when given, else cwd semantics described in --help).

Exit codes (practical)

Code	Typical meaning
`0`	Success (payload may still report logical failures inside JSON for some commands — always parse stdout in scripts).
`1`	Subcommand-specific failure (e.g. `analyze-pr` cannot read diff, graph missing, invalid path for `diagnose-ignore`). For `reprocess`, a requested phase subprocess ran and exited non-zero (see `phases_run` in stdout JSON).
`2`	No subcommand / help printed; `init` refused because the index dir is non-empty; `erase` refused in non-TTY without `--yes`; `meta` when graph payload reports `success: false`; unhandled internal error in `main`. For `reprocess`, invalid flag combination (handled like other argparse errors), or a setup failure before any phase subprocess was spawned (`phases_run: []` in the JSON payload — e.g. cocoindex binary missing next to this Python, flow file missing).

Lifecycle subcommands

`init`

Creates a new index (cocoindex catch-up from empty + full build_ast_graph.py). Refuses if code_graph.lbug or code_index_* Lance tables already exist under the resolved index dir (exit 2).

java-codebase-rag init --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --quiet

`increment`

Runs cocoindex catch-up and incremental LadybugDB graph update. Only changed files and their single-hop dependents are re-parsed and re-written to the graph. Passes 5–6 (client/producer extraction and cross-service matching) run globally. Falls back to full reprocess if:

No previous graph exists (first run)
Graph schema is outdated (missing source_file on edges)
Previous incremental run crashed (crash marker detected)
Dependent expansion exceeds 50 files

java-codebase-rag increment --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --quiet

Flags:

--vectors-only — runs only cocoindex catch-up; skips graph update and emits stale-graph warning. Use this when you want the old Lance-only behavior.

Migration note: After upgrading, run reprocess once to ensure edge tables have source_file columns (ontology version 17+).

`reprocess`

Default (no extra flags): full Lance reprocess (cocoindex --full-reprocess) then full LadybugDB rebuild via build_ast_graph.py, in that order. This remains the recommended coherence operation when both stores might be out of date.

Selective flags (mutually exclusive):

--vectors-only — runs only the cocoindex full reprocess phase; does not invoke the graph builder.
--graph-only — runs only build_ast_graph.py; does not invoke cocoindex.

Passing both flags is rejected by argparse before any subprocess runs. The error is printed on stderr in this form (wording may vary slightly with Python/argparse version):

java-codebase-rag: argument --graph-only: not allowed with argument --vectors-only

Use java-codebase-rag reprocess --help for the live synopsis.

Drift warning (stderr)

After a successful selective run, the CLI prints exactly one line to stderr naming the store that was not rebuilt. --quiet does not suppress this line (quiet only affects subprocess verbosity). There is no extra exit code for drift; scripts should treat stderr as informational.

JSON payload: `phases_run`

The stdout JSON includes an additive list field phases_run: which phases actually spawned subprocesses, in order ("vectors", "graph"). Examples:

Default success after both phases: ["vectors", "graph"]
Default run where cocoindex fails before the graph step: ["vectors"] (graph never started)
--vectors-only success: ["vectors"]
--graph-only success: ["graph"]
Setup failure before any phase (missing cocoindex binary, missing bundled flow file, or pipeline preflight 126/127 stubs): []

Because exit_code and graph_exit_code can be null in multiple situations, prefer branching on phases_run first, then on the relevant per-phase exit field. Asymmetry: --vectors-only reports the cocoindex process in exit_code (and leaves graph_exit_code null); --graph-only leaves top-level exit_code null and reports the graph builder in graph_exit_code, so scripts that only read exit_code miss graph-only outcomes unless they branch on phases_run / graph_exit_code.

java-codebase-rag reprocess --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --quiet
java-codebase-rag reprocess --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --vectors-only --quiet
java-codebase-rag reprocess --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --graph-only --quiet

`erase`

Deletes cocoindex state, the LadybugDB graph (code_graph.lbug), the graph builder's content-hash store (.graph_hashes.json), and Lance tables under the index dir. Requires --yes or interactive confirmation on a TTY. Non-TTY without --yes exits 2.

java-codebase-rag erase --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --yes

Hidden `refresh` alias

java-codebase-rag refresh runs reprocess. Prefer reprocess in scripts.

Introspection subcommands

`meta`

Graph metadata, ontology version, counts, edge_counts, plus resolved embedding fields and provenance (embedding_model_source, embedding_device_source, index_dir, ladybug_path, …).

java-codebase-rag meta --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag

`tables`

Lance table listing and embedding summary (same helper as the server’s table introspection).

java-codebase-rag tables --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag

`diagnose-ignore`

Explains why a path is ignored or not ignored by the layered ignore rules (builtin + project .java-codebase-rag/ignore + nested ignore files + gitignore layers).

java-codebase-rag diagnose-ignore src/main/generated/Foo.java --source-root /path/to/java/repo

`unresolved-calls`

Lists or aggregates receiver-failure call sites stored as UnresolvedCallSite (not on CALLS after ontology 15 PR-3). Reasons: phantom_unresolved_receiver, chained_receiver.

java-codebase-rag unresolved-calls stats --by microservice --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag
java-codebase-rag unresolved-calls list --method-id sym:... --limit 100 --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag

stats --by accepts reason, microservice, or caller_role (declaring type role of the caller method).

Analysis: `analyze-pr`

Maps a unified diff to changed symbols, blast radius, routes touched, and risk band. Requires a built LadybugDB graph at <index-dir>/code_graph.lbug.

Provide exactly one of:

--diff-file PATH
--diff-stdin (read diff from stdin)

git diff > /tmp/pr.diff
java-codebase-rag analyze-pr --diff-file /tmp/pr.diff --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag

Paths in the diff should align with Symbol.filename layout in the graph (project-relative Java paths). Use this from PR-triage scripts or Cursor skills; PR mapping is CLI-only (the MCP exposes retrieval tools only).

Suggested workflows

1. Quick health after a build

java-codebase-rag meta --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag | jq '{ontology_version, parse_errors, counts, edge_counts}'
java-codebase-rag tables --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag | jq '.tables | keys'

2. “Why isn’t this file in the index?”

java-codebase-rag diagnose-ignore path/inside/repo/to/File.java --source-root /path/to/java/repo

3. Full re-index (operator / CI)

java-codebase-rag reprocess --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag --quiet
java-codebase-rag meta --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag | jq .ontology_version

4. PR risk pass (local)

git diff origin/main...HEAD > /tmp/pr.diff
java-codebase-rag analyze-pr --diff-file /tmp/pr.diff --source-root /path/to/java/repo --index-dir /path/to/.java-codebase-rag | jq '{risk_score,risk_band,blast_radius_total}'

Graph-only escape hatch

Prefer java-codebase-rag reprocess --graph-only when you only need LadybugDB rebuilt from the current Lance snapshot. To run the graph builder without going through the CLI (advanced / scripting):

.venv/bin/python build_ast_graph.py --source-root /path/to/java/repo --ladybug-path /path/to/.java-codebase-rag/code_graph.lbug --verbose

`jrag` command — search CLI

The jrag command provides semantic search over the LanceDB index. This is the CLI surface for search (see MCP search tool in AGENT-GUIDE.md for the programmatic interface).

`jrag search`

Semantic search via natural language queries. Returns one row per symbol/type by default; use --chunks to restore chunk-level output.

# Basic search (deduped by default)
jrag search "authentication service"

# Show all chunks (no dedup)
jrag search "authentication service" --chunks

# Hybrid search (vector + keyword)
jrag search "login" --hybrid

# With score breakdown
jrag search "controller" --explain

# With pagination
jrag search "service" --limit 20 --offset 20

Key flags:

--table {java,sql,yaml,all} — Which content table to search (default: java).
--hybrid — Enable vector + keyword hybrid search (single table only).
--explain — Include score breakdown (distance, role weight, symbol bonus).
--chunks — Show every chunk (default collapses to one row per symbol/type).
--limit N — Max hits to return (default 10).
--offset N — Skip N hits (pagination).
--min-score N — Drop hits below this score floor (default 0.0).
--path-contains SUBSTR — Narrow to chunks whose filename contains this substring.
--role ROLE — Filter by role (e.g., CONTROLLER, SERVICE).
--framework FRAMEWORK — Filter by framework (e.g., spring_mvc, webflux).

Breaking change (PR-SEARCH-2): By default, jrag search now returns one row per primary_type_fqn (symbol/type) to prevent a single type from flooding the page. The --chunks flag restores the previous chunk-level output. When deduped, each hit shows a chunks=N field indicating how many chunks were collapsed into that hit.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`java-codebase-rag` CLI — operator guide

Install and discovery

Setup commands

`install`

`update`

Output mode

Indexing progress (stderr)

Environment variables (summary)

Shared flags

Exit codes (practical)

Lifecycle subcommands

`init`

`increment`

`reprocess`

Drift warning (stderr)

JSON payload: `phases_run`

`erase`

Hidden `refresh` alias

Introspection subcommands

`meta`

`tables`

`diagnose-ignore`

`unresolved-calls`

Analysis: `analyze-pr`

Suggested workflows

1. Quick health after a build

2. “Why isn’t this file in the index?”

3. Full re-index (operator / CI)

4. PR risk pass (local)

Graph-only escape hatch

See also

`jrag` command — search CLI

`jrag search`

FilesExpand file tree

JAVA-CODEBASE-RAG-CLI.md

Latest commit

History

JAVA-CODEBASE-RAG-CLI.md

File metadata and controls

java-codebase-rag CLI — operator guide

Install and discovery

Setup commands

install

update

Output mode

Indexing progress (stderr)

Environment variables (summary)

Shared flags

Exit codes (practical)

Lifecycle subcommands

init

increment

reprocess

Drift warning (stderr)

JSON payload: phases_run

erase

Hidden refresh alias

Introspection subcommands

meta

tables

diagnose-ignore

unresolved-calls

Analysis: analyze-pr

Suggested workflows

1. Quick health after a build

2. “Why isn’t this file in the index?”

3. Full re-index (operator / CI)

4. PR risk pass (local)

Graph-only escape hatch

See also

jrag command — search CLI

jrag search

`java-codebase-rag` CLI — operator guide

`install`

`update`

`init`

`increment`

`reprocess`

JSON payload: `phases_run`

`erase`

Hidden `refresh` alias

`meta`

`tables`

`diagnose-ignore`

`unresolved-calls`

Analysis: `analyze-pr`

`jrag` command — search CLI

`jrag search`