patchpal 0.22.4__tar.gz → 0.22.6__tar.gz

{patchpal-0.22.4/patchpal.egg-info → patchpal-0.22.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: patchpal
-Version: 0.22.4
+Version: 0.22.6
 Summary: An agentic coding and automation assistant, supporting both local and cloud LLMs
 Author: PatchPal Contributors
 License-Expression: Apache-2.0
@@ -59,6 +59,7 @@ Most agent frameworks are [built in TypeScript](https://news.ycombinator.com/ite
 
 **Key Features**
 - [Terminal Interface](https://amaiya.github.io/patchpal/usage/interactive/) for interactive development
+- [Sandbox Mode](https://amaiya.github.io/patchpal/usage/sandbox/) for secure container isolation
 - [Python SDK](https://amaiya.github.io/patchpal/usage/python-api/) for flexibility and extensibility
 - [Built-In](https://amaiya.github.io/patchpal/features/tools/) and [Custom Tools](https://amaiya.github.io/patchpal/features/custom-tools/)
 - [Skills System](https://amaiya.github.io/patchpal/features/skills/) and [MCP Integration](https://amaiya.github.io/patchpal/features/mcp/)
@@ -76,7 +77,7 @@ $ pip install patchpal          # install
 $ patchpal --model <model_id>   # start
 ```
 
-> Model support: Any [LiteLLM-supported model](https://models.litellm.ai/) is can be used.
+> Model support: Any [LiteLLM-supported model](https://models.litellm.ai/) can be used.
 > Platform support: Linux, macOS, and Windows are all supported
 
 **Alternative: Run with Docker/Podman (no installation required)**
@@ -173,7 +174,7 @@ While originally designed for software development, PatchPal is also a general-p
 > There are so many coding agent harnesses. Why build yet another one?
 
 1. Most agent harnesses are in TypeScript. We wanted [something in Python](https://amaiya.github.io/patchpal/usage/python-api/) that we could easily extend for our custom workflows.
-2. PatchPal includes a [unique guardrails system](https://amaiya.github.io/patchpal/configuration/#security-permissions) that is better suited to privacy-conscious use cases involving sensitive data.
+2. PatchPal includes a [unique guardrails system](https://amaiya.github.io/patchpal/safety/) that is better suited to privacy-conscious use cases involving sensitive data.
 3. We needed an agent harness that seamlessly works with [both local and cloud models](https://amaiya.github.io/patchpal/models/overview/#supported-models), including AWS GovCloud Bedrock models.
 
 

{patchpal-0.22.4 → patchpal-0.22.6}/README.md

@@ -11,6 +11,7 @@ Most agent frameworks are [built in TypeScript](https://news.ycombinator.com/ite
 
 **Key Features**
 - [Terminal Interface](https://amaiya.github.io/patchpal/usage/interactive/) for interactive development
+- [Sandbox Mode](https://amaiya.github.io/patchpal/usage/sandbox/) for secure container isolation
 - [Python SDK](https://amaiya.github.io/patchpal/usage/python-api/) for flexibility and extensibility
 - [Built-In](https://amaiya.github.io/patchpal/features/tools/) and [Custom Tools](https://amaiya.github.io/patchpal/features/custom-tools/)
 - [Skills System](https://amaiya.github.io/patchpal/features/skills/) and [MCP Integration](https://amaiya.github.io/patchpal/features/mcp/)
@@ -28,7 +29,7 @@ $ pip install patchpal          # install
 $ patchpal --model <model_id>   # start
 ```
 
-> Model support: Any [LiteLLM-supported model](https://models.litellm.ai/) is can be used.
+> Model support: Any [LiteLLM-supported model](https://models.litellm.ai/) can be used.
 > Platform support: Linux, macOS, and Windows are all supported
 
 **Alternative: Run with Docker/Podman (no installation required)**
@@ -125,7 +126,7 @@ While originally designed for software development, PatchPal is also a general-p
 > There are so many coding agent harnesses. Why build yet another one?
 
 1. Most agent harnesses are in TypeScript. We wanted [something in Python](https://amaiya.github.io/patchpal/usage/python-api/) that we could easily extend for our custom workflows.
-2. PatchPal includes a [unique guardrails system](https://amaiya.github.io/patchpal/configuration/#security-permissions) that is better suited to privacy-conscious use cases involving sensitive data.
+2. PatchPal includes a [unique guardrails system](https://amaiya.github.io/patchpal/safety/) that is better suited to privacy-conscious use cases involving sensitive data.
 3. We needed an agent harness that seamlessly works with [both local and cloud models](https://amaiya.github.io/patchpal/models/overview/#supported-models), including AWS GovCloud Bedrock models.
 
 

{patchpal-0.22.4 → patchpal-0.22.6}/patchpal/__init__.py

@@ -1,6 +1,6 @@
 """PatchPal - An open-source Claude Code clone implemented purely in Python."""
 
-__version__ = "0.22.4"
+__version__ = "0.22.6"
 
 from patchpal.agent import create_agent, create_react_agent
 from patchpal.cli.autopilot import autopilot_loop

{patchpal-0.22.4 → patchpal-0.22.6}/patchpal/agent/function_calling.py

@@ -741,6 +741,7 @@ It's currently empty (just the template). The file is automatically loaded at se
                 response = litellm.completion(
                     model=self.model_id,
                     messages=messages,
+                    max_tokens=32000,  # Explicit output token limit for predictable context usage
                     timeout=LLM_TIMEOUT,
                     **self.litellm_kwargs,
                 )
@@ -1078,6 +1079,7 @@ It's currently empty (just the template). The file is automatically loaded at se
                         "messages": messages,
                         "tools": tools,
                         "tool_choice": "auto",
+                        "max_tokens": 32000,  # Explicit output token limit for predictable context usage
                         "timeout": LLM_TIMEOUT,
                         "stream": stream,
                         **self.litellm_kwargs,

{patchpal-0.22.4 → patchpal-0.22.6}/patchpal/cli/sandbox.py

@@ -163,6 +163,7 @@ def detect_llm_endpoints_from_env():
         os.environ.get("AWS_BEDROCK_REGION")
         or os.environ.get("AWS_REGION")
         or os.environ.get("AWS_DEFAULT_REGION")
+        or os.environ.get("AWS_REGION_NAME")  # LiteLLM-specific
     )
     if aws_region and (
         os.environ.get("AWS_ACCESS_KEY_ID") or os.environ.get("AWS_SECRET_ACCESS_KEY")
@@ -427,7 +428,25 @@ def build_container_args(sandbox_args, patchpal_args):
         if needs_privileged:
             print("⚠️  Rootless Podman detected - using --privileged mode for iptables")
             container_args.extend(["--privileged"])
-            container_args.extend(["--cgroup-manager=cgroupfs"])
+            # Only add --cgroup-manager on Linux (not supported on Windows/macOS Podman machines)
+            try:
+                # Check if we're on a Podman machine (macOS/Windows) vs native Linux
+                machine_check = subprocess.run(
+                    ["podman", "machine", "list"], capture_output=True, text=True, check=False
+                )
+                # If machine list works and shows machines, we're on macOS/Windows
+                is_podman_machine = (
+                    machine_check.returncode == 0
+                    and machine_check.stdout.strip()
+                    and len(machine_check.stdout.strip().split("\n")) > 1
+                )
+
+                if not is_podman_machine:
+                    # Native Linux - add cgroup-manager flag
+                    container_args.extend(["--cgroup-manager=cgroupfs"])
+            except Exception:
+                # If check fails, skip the flag (safer default)
+                pass
         else:
             container_args.extend(["--cap-add=NET_ADMIN"])
 
@@ -635,19 +654,22 @@ def build_container_args(sandbox_args, patchpal_args):
     # Add main patchpal command
     if sandbox_args.restrict_network:
         # Use launch-agent-sandbox pattern: model from env var
+        # Falls back to PATCHPAL_MODEL if MODEL not set
         shell_cmd_parts.append(f"""
 echo ""
 echo "=== Starting PatchPal ==="
-echo "Model: $MODEL"
+echo "Model: ${{MODEL:-$PATCHPAL_MODEL}}"
 echo ""
-exec {patchpal_cmd} --model "$MODEL"
+exec {patchpal_cmd} --model "${{MODEL:-$PATCHPAL_MODEL}}"
 """)
     else:
         # Normal mode: use $@ to pass all arguments
+        # Display model from command line or fall back to PATCHPAL_MODEL env var
+        model_display = model_name if model_name else "${PATCHPAL_MODEL:-default}"
         shell_cmd_parts.append(f"""
 echo ""
 echo "=== Starting PatchPal ==="
-echo "Model: {model_name if model_name else "default"}"
+echo "Model: {model_display}"
 echo ""
 exec {patchpal_cmd} "$@"
 """)
@@ -984,6 +1006,31 @@ def main():
         print(f"Loading environment variables from: {sandbox_args.env_file}")
         load_env_file(sandbox_args.env_file)
 
+    # If LITELLM_KWARGS contains AWS params, normalize env var names to uppercase
+    # This must happen before endpoint detection for AWS Bedrock auto-detection to work
+    litellm_kwargs = os.environ.get("PATCHPAL_LITELLM_KWARGS") or os.environ.get(
+        "LITELLM_KWARGS", ""
+    )
+    if litellm_kwargs and "aws_" in litellm_kwargs.lower():
+        # Try to parse as JSON first
+        try:
+            import json
+
+            kwargs_dict = json.loads(litellm_kwargs)
+            # Extract AWS parameters and set as environment variables
+            for key, value in kwargs_dict.items():
+                if key.lower().startswith("aws_"):
+                    upper_key = key.upper()
+                    if upper_key not in os.environ:
+                        os.environ[upper_key] = value
+        except (json.JSONDecodeError, ValueError):
+            # Not JSON, check if individual env vars exist (comma-separated format)
+            for key, value in list(os.environ.items()):
+                if key.lower().startswith("aws_"):
+                    upper_key = key.upper()
+                    if upper_key not in os.environ:
+                        os.environ[upper_key] = value
+
     # Auto-detect LLM endpoints if network restrictions are enabled
     detected_endpoints = []
     if sandbox_args.restrict_network:

{patchpal-0.22.4 → patchpal-0.22.6}/patchpal/context.py

@@ -257,7 +257,10 @@ Be comprehensive but concise. The goal is to continue work seamlessly without lo
         self.system_prompt = system_prompt
         self.estimator = TokenEstimator(model_id)
         self.context_limit = self._get_context_limit()
-        self.output_reserve = 4_096  # Reserve tokens for model output
+        # Reserve 16% of context for output (min 4K, max 32K)
+        # This ensures older models like GPT-4 (8K) get 1.28K reserve
+        # while modern models get full 32K reserve
+        self.output_reserve = min(32_000, max(4_000, int(self.context_limit * 0.16)))
 
     def _get_context_limit(self) -> int:
         """Get context limit for model.

{patchpal-0.22.4 → patchpal-0.22.6/patchpal.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: patchpal
-Version: 0.22.4
+Version: 0.22.6
 Summary: An agentic coding and automation assistant, supporting both local and cloud LLMs
 Author: PatchPal Contributors
 License-Expression: Apache-2.0
@@ -59,6 +59,7 @@ Most agent frameworks are [built in TypeScript](https://news.ycombinator.com/ite
 
 **Key Features**
 - [Terminal Interface](https://amaiya.github.io/patchpal/usage/interactive/) for interactive development
+- [Sandbox Mode](https://amaiya.github.io/patchpal/usage/sandbox/) for secure container isolation
 - [Python SDK](https://amaiya.github.io/patchpal/usage/python-api/) for flexibility and extensibility
 - [Built-In](https://amaiya.github.io/patchpal/features/tools/) and [Custom Tools](https://amaiya.github.io/patchpal/features/custom-tools/)
 - [Skills System](https://amaiya.github.io/patchpal/features/skills/) and [MCP Integration](https://amaiya.github.io/patchpal/features/mcp/)
@@ -76,7 +77,7 @@ $ pip install patchpal          # install
 $ patchpal --model <model_id>   # start
 ```
 
-> Model support: Any [LiteLLM-supported model](https://models.litellm.ai/) is can be used.
+> Model support: Any [LiteLLM-supported model](https://models.litellm.ai/) can be used.
 > Platform support: Linux, macOS, and Windows are all supported
 
 **Alternative: Run with Docker/Podman (no installation required)**
@@ -173,7 +174,7 @@ While originally designed for software development, PatchPal is also a general-p
 > There are so many coding agent harnesses. Why build yet another one?
 
 1. Most agent harnesses are in TypeScript. We wanted [something in Python](https://amaiya.github.io/patchpal/usage/python-api/) that we could easily extend for our custom workflows.
-2. PatchPal includes a [unique guardrails system](https://amaiya.github.io/patchpal/configuration/#security-permissions) that is better suited to privacy-conscious use cases involving sensitive data.
+2. PatchPal includes a [unique guardrails system](https://amaiya.github.io/patchpal/safety/) that is better suited to privacy-conscious use cases involving sensitive data.
 3. We needed an agent harness that seamlessly works with [both local and cloud models](https://amaiya.github.io/patchpal/models/overview/#supported-models), including AWS GovCloud Bedrock models.