Three Capabilities
One API. Three things every agent needs.
Planning
Given a partial result and a goal, Masar produces ordered instructions for what to build next. Dependency-aware, parameterized, ready for your LLM to execute.
Verification
Predicts whether a result is valid in milliseconds, without running the full compiler. Catches 20 categories of structural errors before they become runtime failures.
Memory
Stores agent experiences as structured episodes. Over time, clusters similar experiences into reusable patterns. Your agent gets better at familiar tasks without retraining.
10 Lines to System 2
Your LLM is System 1: fast, intuitive, generates text. Masar is System 2: deliberate, structural, plans and verifies. Wire them together in 10 lines.
from masar import MasarClient
client = MasarClient(api_key="...")
# Plan: what should the agent build next?
plan = client.plan(current=partial, goal=target)
for instruction in plan.instructions:
result = your_llm.execute(instruction)
# Verify: is the result valid?
check = client.verify(schema=result)
print(f"Valid: {check.valid}, Errors: {check.errors}")
# Remember: store this experience
client.memory.store(episode=result, domain="helpdesk")
How It Works
Your agent sends the current state and goal
Masar plans: ordered instructions with exact parameters
Your LLM executes each instruction
Masar verifies: catches errors in milliseconds
Masar remembers: stores the experience for next time
Built For
Agent Builders
Add structured planning and memory to any LangChain, LangGraph, or custom LLM agent. Masar works with any model: Claude, GPT, DeepSeek, open-source.
Orb Developers
Accelerate .orb schema construction with AI-guided planning. Masar knows 129 standard behaviors and produces instructions the compiler can validate.
Enterprise Workflows
Automate business processes with agents that remember. Helpdesk, CRM, compliance workflows that improve from every interaction.
Get Started
Masar is part of the Almadar platform. Join the community.