ctxgraph 0.4.0__tar.gz → 0.5.0__tar.gz

{ctxgraph-0.4.0 → ctxgraph-0.5.0}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: ctxgraph
-Version: 0.4.0
-Summary: 97% cheaper AI coding. ctx init → ctx build → ctx ask. Knowledge graphs, token-efficient capsules, query history, skills, and multi-provider LLM answers (Ollama, Claude, OpenAI, Azure). Cut context costs from cents to fractions of a cent.
+Version: 0.5.0
+Summary: 97% cheaper AI coding. ctxgraph builds a knowledge graph of your Python codebase, then generates token-efficient context capsules for any AI tool. Or hook up Ollama/Claude/OpenAI and use it as a code-aware coding assistant — like ChatGPT or Claude Code, but with built-in graph context. Cut token costs by 97%, work offline with just a graph, or connect your favorite LLM.
 Author: ctxgraph contributors
 License: MIT
 Keywords: code-graph,knowledge-graph,claude,code-analysis,ai-context,token-savings,ollama,openai,ctx,capsule
@@ -29,15 +29,25 @@ Requires-Dist: ruff>=0.1.0; extra == "dev"
 
 # ctxgraph — AI Context Engine for Python
 
-**Slash your LLM token costs by 97%.** Stop dumping entire files into AI context. ctxgraph builds a multi-layer knowledge graph from your Python codebase and generates *compact context capsules* — delivering only what your AI needs, not every line of code.
+## What is ctxgraph?
+
+ctxgraph is a **knowledge graph engine** that makes AI coding assistants **97% cheaper** and **smarter**. 
+
+It analyzes your Python codebase using AST-based static analysis, builds a searchable dependency graph, and delivers *only the code that matters* to your AI — not every line in your project.
+
+### Two ways to use it:
+
+**🔧 As a token-saving context engine** — Output compact DSL capsules for any AI tool (Claude, ChatGPT, Cursor, etc.). Cut context costs from dollars to fractions of a cent.
+
+**🤖 As a full AI coding assistant** — Hook up Ollama (free, local), Claude, or OpenAI, and ask questions, refactor code, debug issues, or explore architecture through `ctx ask` or `ctx chat` — just like ChatGPT or Claude Code, but with built-in codebase awareness.
 
 ```bash
 pip install ctxgraph
 
 ctx init                           # Scaffold .ctxgraph with config + default skills
 ctx build                          # Build knowledge graph (AST analysis → SQLite)
-ctx ask "how does JWT auth work"   # Ask questions with automatic token savings
-ctx capsule "fix JWT expiry"       # 92-99% fewer tokens vs raw code
+ctx ask "how does JWT auth work"   # Ask questions with automatic token savings (needs LLM)
+ctx capsule "fix JWT expiry"       # 92-99% fewer tokens vs raw code (no LLM needed)
 ctx chat "refactor this module"    # Multi-turn conversation with persistent sessions
 ctx history --stats                # Track total tokens saved across all queries
 ccg "fix the login redirect bug"   # Launch Claude with context pre-loaded
@@ -57,11 +67,14 @@ ctx init                           # Creates .ctxgraph/config.toml + default ski
 # 2. Build the knowledge graph
 ctx build                          # AST analysis → SQLite graph
 
-# 3. Ask questions (requires Ollama or other LLM provider)
+# 3a. Ask questions (requires Ollama or other LLM provider — free with Ollama)
 ctx ask "how does authentication work"
 
-# 4. Or generate a capsule for your AI tool
+# 3b. Or just generate a capsule for your AI tool (no LLM needed)
 ctx capsule "fix login rate limiter" --savings
+
+# 4. Explore interactively
+ctx chat                           # Multi-turn REPL with persistent sessions
 ```
 
 ---
@@ -70,6 +83,8 @@ ctx capsule "fix login rate limiter" --savings
 
 Sending entire files to an AI is wasteful. ctxgraph analyzes your code with AST-based static analysis, stores the result in a queryable SQLite graph, and retrieves *only the relevant nodes* — compressed into a token-efficient DSL format.
 
+**No LLM? No problem.** ctxgraph's core features (graph building, capsule generation, search, visualization) work entirely offline. You only need an LLM provider when you want to ask questions via `ctx ask` or chat via `ctx chat`.
+
 | Without ctxgraph | With ctxgraph | Savings |
 |:---|---:|:---:|
 | All files dumped to context | Targeted capsule (10-40 nodes) | **97% fewer tokens** |

{ctxgraph-0.4.0 → ctxgraph-0.5.0}/README.md

@@ -1,14 +1,24 @@
 # ctxgraph — AI Context Engine for Python
 
-**Slash your LLM token costs by 97%.** Stop dumping entire files into AI context. ctxgraph builds a multi-layer knowledge graph from your Python codebase and generates *compact context capsules* — delivering only what your AI needs, not every line of code.
+## What is ctxgraph?
+
+ctxgraph is a **knowledge graph engine** that makes AI coding assistants **97% cheaper** and **smarter**. 
+
+It analyzes your Python codebase using AST-based static analysis, builds a searchable dependency graph, and delivers *only the code that matters* to your AI — not every line in your project.
+
+### Two ways to use it:
+
+**🔧 As a token-saving context engine** — Output compact DSL capsules for any AI tool (Claude, ChatGPT, Cursor, etc.). Cut context costs from dollars to fractions of a cent.
+
+**🤖 As a full AI coding assistant** — Hook up Ollama (free, local), Claude, or OpenAI, and ask questions, refactor code, debug issues, or explore architecture through `ctx ask` or `ctx chat` — just like ChatGPT or Claude Code, but with built-in codebase awareness.
 
 ```bash
 pip install ctxgraph
 
 ctx init                           # Scaffold .ctxgraph with config + default skills
 ctx build                          # Build knowledge graph (AST analysis → SQLite)
-ctx ask "how does JWT auth work"   # Ask questions with automatic token savings
-ctx capsule "fix JWT expiry"       # 92-99% fewer tokens vs raw code
+ctx ask "how does JWT auth work"   # Ask questions with automatic token savings (needs LLM)
+ctx capsule "fix JWT expiry"       # 92-99% fewer tokens vs raw code (no LLM needed)
 ctx chat "refactor this module"    # Multi-turn conversation with persistent sessions
 ctx history --stats                # Track total tokens saved across all queries
 ccg "fix the login redirect bug"   # Launch Claude with context pre-loaded
@@ -28,11 +38,14 @@ ctx init                           # Creates .ctxgraph/config.toml + default ski
 # 2. Build the knowledge graph
 ctx build                          # AST analysis → SQLite graph
 
-# 3. Ask questions (requires Ollama or other LLM provider)
+# 3a. Ask questions (requires Ollama or other LLM provider — free with Ollama)
 ctx ask "how does authentication work"
 
-# 4. Or generate a capsule for your AI tool
+# 3b. Or just generate a capsule for your AI tool (no LLM needed)
 ctx capsule "fix login rate limiter" --savings
+
+# 4. Explore interactively
+ctx chat                           # Multi-turn REPL with persistent sessions
 ```
 
 ---
@@ -41,6 +54,8 @@ ctx capsule "fix login rate limiter" --savings
 
 Sending entire files to an AI is wasteful. ctxgraph analyzes your code with AST-based static analysis, stores the result in a queryable SQLite graph, and retrieves *only the relevant nodes* — compressed into a token-efficient DSL format.
 
+**No LLM? No problem.** ctxgraph's core features (graph building, capsule generation, search, visualization) work entirely offline. You only need an LLM provider when you want to ask questions via `ctx ask` or chat via `ctx chat`.
+
 | Without ctxgraph | With ctxgraph | Savings |
 |:---|---:|:---:|
 | All files dumped to context | Targeted capsule (10-40 nodes) | **97% fewer tokens** |

{ctxgraph-0.4.0 → ctxgraph-0.5.0}/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "ctxgraph"
-version = "0.4.0"
-description = "97% cheaper AI coding. ctx init → ctx build → ctx ask. Knowledge graphs, token-efficient capsules, query history, skills, and multi-provider LLM answers (Ollama, Claude, OpenAI, Azure). Cut context costs from cents to fractions of a cent."
+version = "0.5.0"
+description = "97% cheaper AI coding. ctxgraph builds a knowledge graph of your Python codebase, then generates token-efficient context capsules for any AI tool. Or hook up Ollama/Claude/OpenAI and use it as a code-aware coding assistant — like ChatGPT or Claude Code, but with built-in graph context. Cut token costs by 97%, work offline with just a graph, or connect your favorite LLM."
 readme = "README.md"
 license = {text = "MIT"}
 requires-python = ">=3.10"

{ctxgraph-0.4.0 → ctxgraph-0.5.0}/src/ctxgraph/config/settings.py

@@ -151,11 +151,11 @@ class Settings:
 
     @property
     def temperature(self) -> float:
-        return self._data["ai"].get("temperature", 0.1)
+        return float(self._data["ai"].get("temperature", 0.1))
 
     @property
     def max_tokens(self) -> int:
-        return self._data["ai"].get("max_tokens", 4096)
+        return int(self._data["ai"].get("max_tokens", 4096))
 
     @property
     def exclude_patterns(self) -> list[str]:
@@ -167,11 +167,11 @@ class Settings:
 
     @property
     def max_nodes(self) -> int:
-        return self._data["context"].get("max_nodes", 20)
+        return int(self._data["context"].get("max_nodes", 20))
 
     @property
     def max_depth(self) -> int:
-        return self._data["context"].get("max_depth", 2)
+        return int(self._data["context"].get("max_depth", 2))
 
     @property
     def azure_deployment(self) -> str:
@@ -184,7 +184,7 @@ class Settings:
 
     @property
     def chat_max_session_tokens(self) -> int:
-        return self._data["chat"].get("max_session_tokens", 200000)
+        return int(self._data["chat"].get("max_session_tokens", 200000))
 
     def get_provider_config(self) -> dict:
         return PROVIDER_CONFIGS.get(self.provider, PROVIDER_CONFIGS["custom"])
@@ -218,11 +218,28 @@ class Settings:
                 key, _, value = line.partition("=")
                 key = key.strip()
                 value = value.strip()
-                value = value.strip('"').strip("'")
+
+                if (value.startswith('"') and value.endswith('"')) or \
+                   (value.startswith("'") and value.endswith("'")):
+                    value = value[1:-1]
+                else:
+                    value = Settings._parse_toml_value(value)
+
                 current_section[key] = value
 
         return result
 
+    @staticmethod
+    def _parse_toml_value(value: str):
+        if value.lower() in ("true", "false"):
+            return value.lower() == "true"
+        try:
+            if "." in value:
+                return float(value)
+            return int(value)
+        except ValueError:
+            return value
+
     @staticmethod
     def _deep_merge(base: dict, override: dict):
         for key, value in override.items():

{ctxgraph-0.4.0 → ctxgraph-0.5.0}/src/ctxgraph/mcp/server.py

@@ -44,11 +44,18 @@ from ctxgraph.graph.builder import get_storage
 from ctxgraph.graph.query import search_relevant_nodes
 
 
+def _find_repo_root(start: Path) -> Path:
+    for parent in [start] + list(start.parents):
+        if (parent / ".ctxgraph").is_dir():
+            return parent
+    return start
+
+
 def create_server(repo_path: Optional[str] = None) -> Optional[Any]:
     if not HAS_MCP:
         return None
 
-    path = Path(repo_path).resolve() if repo_path else Path.cwd()
+    path = Path(repo_path).resolve() if repo_path else _find_repo_root(Path.cwd())
     storage = get_storage(path)
 
     server = Server("ctxgraph")

{ctxgraph-0.4.0 → ctxgraph-0.5.0}/src/ctxgraph.egg-info/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: ctxgraph
-Version: 0.4.0
-Summary: 97% cheaper AI coding. ctx init → ctx build → ctx ask. Knowledge graphs, token-efficient capsules, query history, skills, and multi-provider LLM answers (Ollama, Claude, OpenAI, Azure). Cut context costs from cents to fractions of a cent.
+Version: 0.5.0
+Summary: 97% cheaper AI coding. ctxgraph builds a knowledge graph of your Python codebase, then generates token-efficient context capsules for any AI tool. Or hook up Ollama/Claude/OpenAI and use it as a code-aware coding assistant — like ChatGPT or Claude Code, but with built-in graph context. Cut token costs by 97%, work offline with just a graph, or connect your favorite LLM.
 Author: ctxgraph contributors
 License: MIT
 Keywords: code-graph,knowledge-graph,claude,code-analysis,ai-context,token-savings,ollama,openai,ctx,capsule
@@ -29,15 +29,25 @@ Requires-Dist: ruff>=0.1.0; extra == "dev"
 
 # ctxgraph — AI Context Engine for Python
 
-**Slash your LLM token costs by 97%.** Stop dumping entire files into AI context. ctxgraph builds a multi-layer knowledge graph from your Python codebase and generates *compact context capsules* — delivering only what your AI needs, not every line of code.
+## What is ctxgraph?
+
+ctxgraph is a **knowledge graph engine** that makes AI coding assistants **97% cheaper** and **smarter**. 
+
+It analyzes your Python codebase using AST-based static analysis, builds a searchable dependency graph, and delivers *only the code that matters* to your AI — not every line in your project.
+
+### Two ways to use it:
+
+**🔧 As a token-saving context engine** — Output compact DSL capsules for any AI tool (Claude, ChatGPT, Cursor, etc.). Cut context costs from dollars to fractions of a cent.
+
+**🤖 As a full AI coding assistant** — Hook up Ollama (free, local), Claude, or OpenAI, and ask questions, refactor code, debug issues, or explore architecture through `ctx ask` or `ctx chat` — just like ChatGPT or Claude Code, but with built-in codebase awareness.
 
 ```bash
 pip install ctxgraph
 
 ctx init                           # Scaffold .ctxgraph with config + default skills
 ctx build                          # Build knowledge graph (AST analysis → SQLite)
-ctx ask "how does JWT auth work"   # Ask questions with automatic token savings
-ctx capsule "fix JWT expiry"       # 92-99% fewer tokens vs raw code
+ctx ask "how does JWT auth work"   # Ask questions with automatic token savings (needs LLM)
+ctx capsule "fix JWT expiry"       # 92-99% fewer tokens vs raw code (no LLM needed)
 ctx chat "refactor this module"    # Multi-turn conversation with persistent sessions
 ctx history --stats                # Track total tokens saved across all queries
 ccg "fix the login redirect bug"   # Launch Claude with context pre-loaded
@@ -57,11 +67,14 @@ ctx init                           # Creates .ctxgraph/config.toml + default ski
 # 2. Build the knowledge graph
 ctx build                          # AST analysis → SQLite graph
 
-# 3. Ask questions (requires Ollama or other LLM provider)
+# 3a. Ask questions (requires Ollama or other LLM provider — free with Ollama)
 ctx ask "how does authentication work"
 
-# 4. Or generate a capsule for your AI tool
+# 3b. Or just generate a capsule for your AI tool (no LLM needed)
 ctx capsule "fix login rate limiter" --savings
+
+# 4. Explore interactively
+ctx chat                           # Multi-turn REPL with persistent sessions
 ```
 
 ---
@@ -70,6 +83,8 @@ ctx capsule "fix login rate limiter" --savings
 
 Sending entire files to an AI is wasteful. ctxgraph analyzes your code with AST-based static analysis, stores the result in a queryable SQLite graph, and retrieves *only the relevant nodes* — compressed into a token-efficient DSL format.
 
+**No LLM? No problem.** ctxgraph's core features (graph building, capsule generation, search, visualization) work entirely offline. You only need an LLM provider when you want to ask questions via `ctx ask` or chat via `ctx chat`.
+
 | Without ctxgraph | With ctxgraph | Savings |
 |:---|---:|:---:|
 | All files dumped to context | Targeted capsule (10-40 nodes) | **97% fewer tokens** |