Skip to main content

Introduction to SLOP

SLOP (Structured Language for Orchestrating Prompts) is a sandboxed execution environment for LLM-generated code.

Scripts run under hard limits on iterations, LLM calls, API calls, duration, and cost. Any script can pause mid-run — and when it does, the entire execution state is written to a plain JSON checkpoint: the source code, the pause position, every variable in every scope, the call stack, and all output emitted so far.

That state is editable. When a generated script makes a bad API call three MCP calls into a workflow, the checkpoint gives you full freedom over how to continue:

  • Skip the call — move position.statement_index past the failing statement.
  • Patch the result — make the call yourself and paste the real response into the variable the script would have written, or drop in a placeholder and keep moving.
  • Rewrite the script — fix the call, or patch the loop's try/catch so one failure stops killing the batch. Update script_hash (SHA-256 of the new source) and resume.
slop run agent.slop --checkpoint-dir ./checkpoints
# Script paused. Checkpoint saved to: ./checkpoints/20260801_193635.json

# ... a later API call fails on resume; the checkpoint is still on disk ...
# ... edit the checkpoint JSON — code, position, or variables ...

slop resume ./checkpoints/patched.json
# continues from the pause point, completed work restored

What is SLOP?​

SLOP is a Python-like scripting language whose runtime treats execution state as data, optimized for:

  • 🤖 Building AI agents and chatbots
  • 🔄 Orchestrating LLM workflows
  • 🔌 Chaining MCP tool calls and external services
  • 🛠️ Prompt engineering and testing
  • 📊 Data processing for AI applications

Why SLOP?​

Pause, Edit, Resume​

repos = github.search(query: "mcp servers")
pause("after_fetch")   # full execution state saved as editable JSON
result = llm.call(prompt: "Summarize: " + json_stringify(repos), schema: {summary: string})
emit(result.summary)

A pause writes the whole runtime — code, stack, memory, emitted output — to a checkpoint file. Edit any of it, then slop resume continues from that exact point.

Sandboxed by Default​

  • 🧱 Pure language — no filesystem, network, shell, or imports; scripts only touch host-registered services
  • ⏱️ Automatic timeout protection
  • 🔁 Loop iteration limits
  • 📏 LLM call, API call, cost, and call-depth limits
  • 🛡️ Schema validation

Native AI Features​

  • LLM Integration: Call language models with llm.call()
  • MCP Support: Connect MCP servers and call their tools as services
  • Streaming: Real-time output with emit statements
  • Modules: Organize agents into reusable components

Quick Example​

Here's a complete AI assistant in SLOP:

# Process a question with an LLM
user_msg = "What is SLOP?"

response = llm.call(prompt: user_msg, schema: {answer: string})

# Stream the response
emit(response.answer)

What Makes SLOP Different?​

FeatureSLOPPythonJavaScript
Pausable, Editable Execution✅ JSON checkpoints + resume❌❌
AI-First Design✅ Native LLM calls❌ Requires libraries❌ Requires libraries
Safety Built-in✅ Automatic limits⚠️ Manual⚠️ Manual
Streaming Support✅ emit keyword❌ Complex❌ Complex
Learning Curve🟢 5 minutes🟡 Hours🟡 Hours
MCP Integration✅ Native❌ Manual❌ Manual

Next Steps​

Install SLOP →

Your First Script →

View Examples →

Community & Support​

Ready to build your first AI agent? Let's get started! 🚀