Skip to content

Zubbl SDK

Make any AI agent 20-50% smarter automatically.

Drop-in SDK that adds trajectory learning, failure recovery, and continuous improvement to any agent framework. No infrastructure to manage.

Quick Start

pip install zubbl-sdk

from zubbl import ZubblClient

zubbl = ZubblClient(api_key="zubbl_xxx")
smart_agent = zubbl.wrap(your_agent)
result = smart_agent.run("Complete this task")

That's it. Your agent now learns from every execution, recovers from failures, and improves over time.

How It Works

graph LR
    A[Your Agent] --> B[Zubbl SDK]
    B --> C[Execute Task]
    C --> D{Success?}
    D -->|Yes| E[Learn Pattern]
    D -->|No| F[Reflect + Retry]
    F --> C
    E --> G[Improve Policies]
    G --> A
  1. Record — SDK captures every step your agent takes
  2. Learn — Patterns are extracted and scored automatically
  3. Improve — Next run gets better recommendations
  4. Recover — Failed tasks are retried with learned insights

Core Features

Feature What It Does
Trajectory Recording Auto-captures steps, tools, tokens, latency
Policy Recommendations Get optimized strategies before each task
Failure Recovery ~33% of failures auto-recovered on first retry
Improvement Tracking Measure exactly how much your agents improved

Cloud Features (Pro plans)

Feature What It Does
Skill Graph Map skill dependencies and find gaps
Vector Search Find similar past trajectories semantically
Training Continuous policy optimization
Real-time Streams Live monitoring of agent activity

Plans

Plan Price Credits/mo Features
Free $0 10,000 Core SDK, trajectory learning, recovery
Local Pro $29/mo 100,000 + APO prompt optimization, guardrails
Cloud Pro $99/mo 500,000 + Skill graph, vector search, training
Team $299/mo 2,000,000 + Multi-tenant, SSO, dedicated support