code-context-engine 0.4.5__tar.gz → 0.4.6__tar.gz

{code_context_engine-0.4.5/src/code_context_engine.egg-info → code_context_engine-0.4.6}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: code-context-engine
-Version: 0.4.5
-Summary: Index your codebase. AI searches instead of re-reading files. 93% token savings, benchmarked on FastAPI. Works with Claude Code, Cursor, VS Code, Gemini CLI, and Codex.
+Version: 0.4.6
+Summary: Index your codebase. AI searches instead of re-reading files. 94% token savings, benchmarked on FastAPI. Works with Claude Code, Cursor, VS Code, Gemini CLI, and Codex.
 Author-email: Fazle Elahee <felahee@gmail.com>, Raj <rajkumar.sakti@gmail.com>
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/elara-labs/code-context-engine
@@ -54,7 +54,7 @@ Dynamic: license-file
 <h1 align="center">Code Context Engine</h1>
 
 <p align="center">
-  <strong>Index your codebase. AI searches instead of re-reading files. 93% token savings, benchmarked.</strong>
+  <strong>Index your codebase. AI searches instead of re-reading files. 94% token savings, benchmarked.</strong>
 </p>
 
 <p align="center">
@@ -122,7 +122,7 @@ Multiple editors in the same project? All get configured in one command.
 ```
   my-project · 38 queries
 
-  ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶  93% tokens saved
+  ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶  94% tokens saved
 
   Without CCE   48.0k  tokens   $0.24
   With CCE       3.4k  tokens   $0.02
@@ -136,7 +136,7 @@ Multiple editors in the same project? All get configured in one command.
 
 ## Why this matters
 
-Input tokens are 85-95% of your Claude Code bill. CCE cuts them by 93% ([benchmarked on FastAPI](#benchmark-fastapi-independently-verified)).
+Input tokens are 85-95% of your Claude Code bill. CCE cuts them by 94% ([benchmarked on FastAPI](#benchmark-fastapi-independently-verified)).
 
 ```
 Without CCE:    Claude reads payments.py + shipping.py   = 45,000 tokens
@@ -154,17 +154,16 @@ With CCE:       context_search "payment flow"            =    800 tokens
 
 ## Benchmark: FastAPI (independently verified)
 
-We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (48 source files, 19K lines of Python) with 20 real coding questions. No cherry-picking, no synthetic queries.
+We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (53 source files, 180K tokens) with 20 real coding questions. No cherry-picking, no synthetic queries.
 
 **Methodology:** For each query, "without CCE" means reading the full content of every file the query touches. "With CCE" means the relevant chunks after compression. This is conservative (agents often read more files than needed).
 
 | Metric | Result |
 |--------|--------|
-| **Retrieval** | **93%** savings (75,355 → 5,381 tokens/query) |
-| **+ Compression** | **90%** additional (5,381 → 541 tokens/query) |
-| **Combined** | **99.3%** (75,355 → 541 tokens/query) |
-| Recall@10 (found the right files) | 0.80 |
-| Precision@10 | 0.30 |
+| **Retrieval** | **94%** savings (83,681 → 4,927 tokens/query) |
+| **+ Compression** | **89%** additional (4,927 → 523 tokens/query) |
+| **Combined** | **99.4%** (83,681 → 523 tokens/query) |
+| Recall@10 (found the right files) | 0.90 |
 | Latency p50 | 0.4ms |
 | Queries tested | 20 |
 
@@ -172,8 +171,8 @@ We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (48 sou
 
 | Layer | What it does | Savings | Method |
 |-------|-------------|---------|--------|
-| **Retrieval** | Full files → relevant code chunks | 93% | measured |
-| **Chunk Compression** | Raw chunks → signatures + docstrings | 90% | measured |
+| **Retrieval** | Full files → relevant code chunks | 94% | measured |
+| **Chunk Compression** | Raw chunks → signatures + docstrings | 89% | measured |
 | **Output Compression** | Reduces Claude's reply length | 65% | estimated |
 | **Grammar** | Drops articles/fillers from memory text | 13% | measured |
 
@@ -246,7 +245,7 @@ Re-indexing after edits takes under 1 second (96% embedding cache hit rate). Git
 
 Output compression tools (like Caveman) save 20-75% on output tokens. Output is 5-15% of your bill. Net savings: ~11%.
 
-CCE saves on **input** tokens (93% retrieval + 90% compression on FastAPI, [independently benchmarked](#benchmark-fastapi-independently-verified)). Input is 85-95% of your bill.
+CCE saves on **input** tokens (94% retrieval + 89% compression on FastAPI, [independently benchmarked](#benchmark-fastapi-independently-verified)). Input is 85-95% of your bill.
 
 ### It actually understands your code
 
@@ -416,7 +415,7 @@ No GPU required. Embedding model runs on CPU via ONNX Runtime.
 - [x] Clean uninstall (removes all CCE artifacts)
 - [x] AST-aware chunking for PHP, Go, Rust, Java (tree-sitter)
 - [x] Multi-editor support (Cursor, VS Code/Copilot, Gemini CLI)
-- [x] Reproducible benchmark suite (93% savings on FastAPI, per-layer breakdown)
+- [x] Reproducible benchmark suite (94% savings on FastAPI, per-layer breakdown)
 - [x] Session savings visibility (shown at every session start)
 - [ ] Tree-sitter support for C, C++, Ruby, Swift, Kotlin
 - [ ] Docker support for remote mode

{code_context_engine-0.4.5 → code_context_engine-0.4.6}/README.md

@@ -5,7 +5,7 @@
 <h1 align="center">Code Context Engine</h1>
 
 <p align="center">
-  <strong>Index your codebase. AI searches instead of re-reading files. 93% token savings, benchmarked.</strong>
+  <strong>Index your codebase. AI searches instead of re-reading files. 94% token savings, benchmarked.</strong>
 </p>
 
 <p align="center">
@@ -73,7 +73,7 @@ Multiple editors in the same project? All get configured in one command.
 ```
   my-project · 38 queries
 
-  ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶  93% tokens saved
+  ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶  94% tokens saved
 
   Without CCE   48.0k  tokens   $0.24
   With CCE       3.4k  tokens   $0.02
@@ -87,7 +87,7 @@ Multiple editors in the same project? All get configured in one command.
 
 ## Why this matters
 
-Input tokens are 85-95% of your Claude Code bill. CCE cuts them by 93% ([benchmarked on FastAPI](#benchmark-fastapi-independently-verified)).
+Input tokens are 85-95% of your Claude Code bill. CCE cuts them by 94% ([benchmarked on FastAPI](#benchmark-fastapi-independently-verified)).
 
 ```
 Without CCE:    Claude reads payments.py + shipping.py   = 45,000 tokens
@@ -105,17 +105,16 @@ With CCE:       context_search "payment flow"            =    800 tokens
 
 ## Benchmark: FastAPI (independently verified)
 
-We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (48 source files, 19K lines of Python) with 20 real coding questions. No cherry-picking, no synthetic queries.
+We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (53 source files, 180K tokens) with 20 real coding questions. No cherry-picking, no synthetic queries.
 
 **Methodology:** For each query, "without CCE" means reading the full content of every file the query touches. "With CCE" means the relevant chunks after compression. This is conservative (agents often read more files than needed).
 
 | Metric | Result |
 |--------|--------|
-| **Retrieval** | **93%** savings (75,355 → 5,381 tokens/query) |
-| **+ Compression** | **90%** additional (5,381 → 541 tokens/query) |
-| **Combined** | **99.3%** (75,355 → 541 tokens/query) |
-| Recall@10 (found the right files) | 0.80 |
-| Precision@10 | 0.30 |
+| **Retrieval** | **94%** savings (83,681 → 4,927 tokens/query) |
+| **+ Compression** | **89%** additional (4,927 → 523 tokens/query) |
+| **Combined** | **99.4%** (83,681 → 523 tokens/query) |
+| Recall@10 (found the right files) | 0.90 |
 | Latency p50 | 0.4ms |
 | Queries tested | 20 |
 
@@ -123,8 +122,8 @@ We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (48 sou
 
 | Layer | What it does | Savings | Method |
 |-------|-------------|---------|--------|
-| **Retrieval** | Full files → relevant code chunks | 93% | measured |
-| **Chunk Compression** | Raw chunks → signatures + docstrings | 90% | measured |
+| **Retrieval** | Full files → relevant code chunks | 94% | measured |
+| **Chunk Compression** | Raw chunks → signatures + docstrings | 89% | measured |
 | **Output Compression** | Reduces Claude's reply length | 65% | estimated |
 | **Grammar** | Drops articles/fillers from memory text | 13% | measured |
 
@@ -197,7 +196,7 @@ Re-indexing after edits takes under 1 second (96% embedding cache hit rate). Git
 
 Output compression tools (like Caveman) save 20-75% on output tokens. Output is 5-15% of your bill. Net savings: ~11%.
 
-CCE saves on **input** tokens (93% retrieval + 90% compression on FastAPI, [independently benchmarked](#benchmark-fastapi-independently-verified)). Input is 85-95% of your bill.
+CCE saves on **input** tokens (94% retrieval + 89% compression on FastAPI, [independently benchmarked](#benchmark-fastapi-independently-verified)). Input is 85-95% of your bill.
 
 ### It actually understands your code
 
@@ -367,7 +366,7 @@ No GPU required. Embedding model runs on CPU via ONNX Runtime.
 - [x] Clean uninstall (removes all CCE artifacts)
 - [x] AST-aware chunking for PHP, Go, Rust, Java (tree-sitter)
 - [x] Multi-editor support (Cursor, VS Code/Copilot, Gemini CLI)
-- [x] Reproducible benchmark suite (93% savings on FastAPI, per-layer breakdown)
+- [x] Reproducible benchmark suite (94% savings on FastAPI, per-layer breakdown)
 - [x] Session savings visibility (shown at every session start)
 - [ ] Tree-sitter support for C, C++, Ruby, Swift, Kotlin
 - [ ] Docker support for remote mode

{code_context_engine-0.4.5 → code_context_engine-0.4.6}/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "code-context-engine"
-version = "0.4.5"
-description = "Index your codebase. AI searches instead of re-reading files. 93% token savings, benchmarked on FastAPI. Works with Claude Code, Cursor, VS Code, Gemini CLI, and Codex."
+version = "0.4.6"
+description = "Index your codebase. AI searches instead of re-reading files. 94% token savings, benchmarked on FastAPI. Works with Claude Code, Cursor, VS Code, Gemini CLI, and Codex."
 readme = {file = "README.md", content-type = "text/markdown"}
 license = "MIT"
 authors = [

{code_context_engine-0.4.5 → code_context_engine-0.4.6/src/code_context_engine.egg-info}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: code-context-engine
-Version: 0.4.5
-Summary: Index your codebase. AI searches instead of re-reading files. 93% token savings, benchmarked on FastAPI. Works with Claude Code, Cursor, VS Code, Gemini CLI, and Codex.
+Version: 0.4.6
+Summary: Index your codebase. AI searches instead of re-reading files. 94% token savings, benchmarked on FastAPI. Works with Claude Code, Cursor, VS Code, Gemini CLI, and Codex.
 Author-email: Fazle Elahee <felahee@gmail.com>, Raj <rajkumar.sakti@gmail.com>
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/elara-labs/code-context-engine
@@ -54,7 +54,7 @@ Dynamic: license-file
 <h1 align="center">Code Context Engine</h1>
 
 <p align="center">
-  <strong>Index your codebase. AI searches instead of re-reading files. 93% token savings, benchmarked.</strong>
+  <strong>Index your codebase. AI searches instead of re-reading files. 94% token savings, benchmarked.</strong>
 </p>
 
 <p align="center">
@@ -122,7 +122,7 @@ Multiple editors in the same project? All get configured in one command.
 ```
   my-project · 38 queries
 
-  ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶  93% tokens saved
+  ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶  94% tokens saved
 
   Without CCE   48.0k  tokens   $0.24
   With CCE       3.4k  tokens   $0.02
@@ -136,7 +136,7 @@ Multiple editors in the same project? All get configured in one command.
 
 ## Why this matters
 
-Input tokens are 85-95% of your Claude Code bill. CCE cuts them by 93% ([benchmarked on FastAPI](#benchmark-fastapi-independently-verified)).
+Input tokens are 85-95% of your Claude Code bill. CCE cuts them by 94% ([benchmarked on FastAPI](#benchmark-fastapi-independently-verified)).
 
 ```
 Without CCE:    Claude reads payments.py + shipping.py   = 45,000 tokens
@@ -154,17 +154,16 @@ With CCE:       context_search "payment flow"            =    800 tokens
 
 ## Benchmark: FastAPI (independently verified)
 
-We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (48 source files, 19K lines of Python) with 20 real coding questions. No cherry-picking, no synthetic queries.
+We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (53 source files, 180K tokens) with 20 real coding questions. No cherry-picking, no synthetic queries.
 
 **Methodology:** For each query, "without CCE" means reading the full content of every file the query touches. "With CCE" means the relevant chunks after compression. This is conservative (agents often read more files than needed).
 
 | Metric | Result |
 |--------|--------|
-| **Retrieval** | **93%** savings (75,355 → 5,381 tokens/query) |
-| **+ Compression** | **90%** additional (5,381 → 541 tokens/query) |
-| **Combined** | **99.3%** (75,355 → 541 tokens/query) |
-| Recall@10 (found the right files) | 0.80 |
-| Precision@10 | 0.30 |
+| **Retrieval** | **94%** savings (83,681 → 4,927 tokens/query) |
+| **+ Compression** | **89%** additional (4,927 → 523 tokens/query) |
+| **Combined** | **99.4%** (83,681 → 523 tokens/query) |
+| Recall@10 (found the right files) | 0.90 |
 | Latency p50 | 0.4ms |
 | Queries tested | 20 |
 
@@ -172,8 +171,8 @@ We benchmarked CCE against [FastAPI](https://github.com/fastapi/fastapi) (48 sou
 
 | Layer | What it does | Savings | Method |
 |-------|-------------|---------|--------|
-| **Retrieval** | Full files → relevant code chunks | 93% | measured |
-| **Chunk Compression** | Raw chunks → signatures + docstrings | 90% | measured |
+| **Retrieval** | Full files → relevant code chunks | 94% | measured |
+| **Chunk Compression** | Raw chunks → signatures + docstrings | 89% | measured |
 | **Output Compression** | Reduces Claude's reply length | 65% | estimated |
 | **Grammar** | Drops articles/fillers from memory text | 13% | measured |
 
@@ -246,7 +245,7 @@ Re-indexing after edits takes under 1 second (96% embedding cache hit rate). Git
 
 Output compression tools (like Caveman) save 20-75% on output tokens. Output is 5-15% of your bill. Net savings: ~11%.
 
-CCE saves on **input** tokens (93% retrieval + 90% compression on FastAPI, [independently benchmarked](#benchmark-fastapi-independently-verified)). Input is 85-95% of your bill.
+CCE saves on **input** tokens (94% retrieval + 89% compression on FastAPI, [independently benchmarked](#benchmark-fastapi-independently-verified)). Input is 85-95% of your bill.
 
 ### It actually understands your code
 
@@ -416,7 +415,7 @@ No GPU required. Embedding model runs on CPU via ONNX Runtime.
 - [x] Clean uninstall (removes all CCE artifacts)
 - [x] AST-aware chunking for PHP, Go, Rust, Java (tree-sitter)
 - [x] Multi-editor support (Cursor, VS Code/Copilot, Gemini CLI)
-- [x] Reproducible benchmark suite (93% savings on FastAPI, per-layer breakdown)
+- [x] Reproducible benchmark suite (94% savings on FastAPI, per-layer breakdown)
 - [x] Session savings visibility (shown at every session start)
 - [ ] Tree-sitter support for C, C++, Ruby, Swift, Kotlin
 - [ ] Docker support for remote mode

{code_context_engine-0.4.5 → code_context_engine-0.4.6}/src/context_engine/cli.py

@@ -1674,12 +1674,10 @@ def search(ctx: click.Context, query: str, top_k: int) -> None:
             lines.append(f"    {DOT} {dim('No results found')}")
         else:
             # Compute tokens
-            raw_tokens = 0
             served_tokens = 0
             seen_files: set[str] = set()
             for r in results:
                 chunk_tokens = max(1, len(r.content) // 4)
-                raw_tokens += chunk_tokens
                 served_tokens += chunk_tokens
                 seen_files.add(r.file_path)
 
@@ -1702,7 +1700,8 @@ def search(ctx: click.Context, query: str, top_k: int) -> None:
                 lines.append(f"       {dim(first_line)}")
 
             lines.append("")
-            lines.append(f"    {CHECK} {success(f'{len(results)} results')}  {dim(f'{served_tokens} tokens served vs {full_file_tokens} full file tokens')}")
+            savings_pct = int((1 - served_tokens / full_file_tokens) * 100) if full_file_tokens > 0 else 0
+            lines.append(f"    {CHECK} {success(f'{len(results)} results')}  {dim(f'{served_tokens} tokens served vs {full_file_tokens} full file tokens ({savings_pct}% saved)')}")
 
             # Update stats
             stats_path = storage_dir / "stats.json"
@@ -1711,10 +1710,8 @@ def search(ctx: click.Context, query: str, top_k: int) -> None:
             except (json.JSONDecodeError, OSError):
                 stats = {}
             stats["queries"] = stats.get("queries", 0) + 1
-            stats["raw_tokens"] = stats.get("raw_tokens", 0) + raw_tokens
+            stats["full_file_tokens"] = stats.get("full_file_tokens", 0) + full_file_tokens
             stats["served_tokens"] = stats.get("served_tokens", 0) + served_tokens
-            stats.setdefault("full_file_tokens", 0)
-            stats["full_file_tokens"] = max(stats["full_file_tokens"], full_file_tokens)
             stats_path.write_text(json.dumps(stats))
 
         lines.append("")
@@ -1724,12 +1721,19 @@ def search(ctx: click.Context, query: str, top_k: int) -> None:
 
 
 @main.command()
-def uninstall() -> None:
+@click.option("--yes", "-y", is_flag=True, help="Skip confirmation prompt")
+def uninstall(yes: bool) -> None:
     """Remove CCE from the current project (hooks, .mcp.json entry, CLAUDE.md block)."""
     from context_engine.cli_style import section, animate, value, dim, success, warn, CHECK, CROSS, DOT
 
     project_dir = Path.cwd()
     project_name = project_dir.name
+
+    if not yes:
+        if not click.confirm(f"Remove CCE from {project_name}?", default=False):
+            click.echo("Cancelled.")
+            return
+
     lines: list[str] = []
     lines.append("")
     lines.append(section(f"Uninstall · {project_name}"))

{code_context_engine-0.4.5 → code_context_engine-0.4.6}/src/context_engine/retrieval/confidence.py

@@ -14,8 +14,8 @@ import time
 from context_engine.models import Chunk
 
 _VECTOR_WEIGHT = 0.5
-_KEYWORD_WEIGHT = 0.3
-_RECENCY_WEIGHT = 0.2
+_KEYWORD_WEIGHT = 0.4
+_RECENCY_WEIGHT = 0.1
 _MAX_KEYWORD_DISTANCE = 5
 _RECENCY_HALF_LIFE = 7 * 24 * 3600  # 1 week
 

{code_context_engine-0.4.5 → code_context_engine-0.4.6}/src/context_engine/retrieval/retriever.py

@@ -15,6 +15,9 @@ _RRF_K = 60
 # [0,1] by the best score in the candidate set so an exact-match FTS rank-1 hit
 # scores the same as a vector rank-1 hit instead of being clamped to ~1.0.
 _CONFIDENCE_WEIGHT = 0.5
+# Max chunks from the same file in the final result set. Prevents one large
+# file from dominating results and improves file-level precision.
+_MAX_CHUNKS_PER_FILE = 3
 # When the parsed query looks like a code lookup, give FTS more pull because
 # exact-identifier hits are usually what the user wants.
 _FTS_BOOST_CODE_LOOKUP = 1.5
@@ -129,7 +132,20 @@ class HybridRetriever:
                 scored.append((chunk, final_score))
 
         scored.sort(key=lambda x: x[1], reverse=True)
-        ranked = [chunk for chunk, _ in scored[:top_k]]
+
+        # File diversity: cap chunks per file so one large file doesn't
+        # dominate the result set. This improves precision by letting
+        # chunks from more files surface into the top-k.
+        file_counts: dict[str, int] = {}
+        diverse: list[Chunk] = []
+        for chunk, _ in scored:
+            count = file_counts.get(chunk.file_path, 0)
+            if count < _MAX_CHUNKS_PER_FILE:
+                diverse.append(chunk)
+                file_counts[chunk.file_path] = count + 1
+                if len(diverse) >= top_k:
+                    break
+        ranked = diverse
 
         # Graph expansion: fetch 1-2 bonus chunks from files reachable via
         # CALLS/IMPORTS edges from the top results.

{code_context_engine-0.4.5 → code_context_engine-0.4.6}/tests/test_cli_uninstall.py

@@ -31,7 +31,7 @@ def _run_uninstall_in(runner, project_dir: Path):
     original = Path.cwd()
     try:
         os.chdir(project_dir)
-        return runner.invoke(main, ["uninstall"])
+        return runner.invoke(main, ["uninstall", "--yes"])
     finally:
         os.chdir(original)