Skip to content

CodeNib Names And Paths

CodeNib is both the product name and the maintained Python and command-line namespace. Current installations use the codenib package, codenib command, CODENIB_* environment variables, and CodeNib-owned state paths.

Preferred Commands

New installations should use the unified product command:

codenib wiki /path/to/repo
codenib mcp /path/to/repo

Python uses the codenib distribution and import package:

from codenib.agent import CodeNibAgentOptions

Stable Inputs

The following identifiers are stable public inputs:

  • CODENIB_* environment variables;
  • ~/.codenib and .codenib_qa state roots;
  • the codenib package and every existing codenib-* command;
  • MCP server ID codenib, prompt ID codenib-guide, tool names, and codenib_context skill ID;
  • manifest, graph, vector, and incremental-cache formats, including the repo_manifest.json, graph.pkl, incremental_state.json, chunk_store.pkl, embeddings_cache.pkl, and qa_registry.json entry filenames;
  • canonical repository URL https://github.com/sysevol-ai/CodeNib;
  • CodeNib Base, CodeNib Synthesis, their Hub IDs, and artifact paths;
  • public classes such as CodeNibAgentOptions, CodeNibBaseDataset, and CodeNibSynthesisDataset.

The Hub values above are external dataset addresses, not aliases for executable CodeNib interfaces. Persisted schemas change only when their serialized structure or declared identity changes.

Filesystem Roots

CodeNib resolves machine-dependent locations through four environment variables:

Variable Default Purpose
CODENIB_HOME ~/.codenib User-owned datasets, checkouts, and state
CODENIB_PREBUILT_DIR $CODENIB_HOME/prebuilt Reusable repository artifacts
CODENIB_RESULTS_DIR $CODENIB_HOME/results Benchmark and experiment outputs
CODENIB_TEMP_DIR $TMPDIR/codenib Disposable indexer and tool work

User-facing commands keep repository indexes under $CODENIB_HOME/repositories/<repo>-<id>/indexes (default ~/.codenib/repositories/...) instead of storing CodeNib artifacts in the target checkout. Language-aware graph builders may still invoke project toolchains that prepare dependencies in the checkout. The lower-level compiler accepts an explicit cache_dir; compatibility code can also read older repository-local .codenib_cache manifests.