Runner Targets Overview

Overview

Connect your agent code to FluxLoop by configuring the runner section in configs/simulation.yaml. Multiple integration patterns support different languages, frameworks, and execution environments.

Pattern Categories

🟢 P0: Production-Ready

Pattern	When to Use	Documentation
Python Function/Method	Direct sync/async function calls	python-function
Python Class	Class instance methods (with factory support)	Coming soon
Python Async Generator	Streaming responses (OpenAI/Anthropic SDKs)	Coming soon
HTTP REST/SSE	Remote APIs or local servers (with streaming)	http-sse
WebSocket	Bidirectional streaming (real-time chat)	Coming soon
Subprocess (JSONL)	Cross-language runtimes (Node/Go/etc)	subprocess-jsonl
Step Loop	Multi-turn conversations (step() iterations)	Coming soon
Resource Guards	Timeout/output limits safety net	Coming soon

🟡 P1: Beta Features

Pattern	When to Use	Documentation
Batch Execution	Parallel evaluation on JSONL/CSV datasets	Coming soon
Advanced Streaming	Custom event paths for tool/function calls	Coming soon
Input/Output Adapters	Signature transformations between frameworks	Coming soon

🔴 P2: Experimental (Roadmap)

Pattern	When to Use	Documentation
Docker Container	Isolated, reproducible execution	Coming soon
Redis/SQS Queue	Production-scale async workers	Coming soon
Multi-Target Composite	Sequential/parallel/ensemble agents	Coming soon

Quick Start

1) Python Function (Simplest)

runner:
  target: "app.agent:run"
  working_directory: .

2) HTTP API (Remote Service)

runner:
  http:
    method: POST
    url: "http://localhost:8000/chat"
    stream: sse

3) Subprocess (Node.js Agent)

runner:
  process:
    command: ["node", "agent.mjs"]
    protocol: jsonl

Common Options

Available across all runners:

runner:
  # ... pattern-specific config ...
  
  # Common options
  working_directory: .
  python_path: ["src", "lib"]           # Python patterns only
  stream_output_path: "message.delta"   # Streaming runners
  
  # Resource guards (optional)
  guards:
    max_duration: 120s
    output_char_limit: 20000

Argument Replay (Optional)

Reuse complex kwargs:

runner:
  target: "app.agent:run"
  
replay_args:
  enabled: true
  recording_file: recordings/args_recording.jsonl
  # override_param_path: "item.content.0.text"

See simulation configuration for details.

Next Steps

• Explore pattern-specific docs for full options and examples
• Simulation Config for complete structure
• Run Command for CLI usage

MCP Server Integration

These docs serve as FluxLoop MCP server's knowledge base:

• analyze_repository → framework detection → pattern recommendations
• generate_integration_steps → pattern-specific checklists
• faq → troubleshooting/example search across all patterns

MCP Server Plan: docs/prd/mcp_server_plan.md