PenguinHarnessYour Automated Agent Builder, right on the Web
Create self-evolving agents online in one click — no install, no setup. PenguinHarness builds, evaluates and improves agents for you, connected to 1,000+ models.
Open the online console, type one sentence, and watch agents build agents — at 100× speed.
- supported models
- 1000+
- to build an agent app
- $0.02
- open source
- 100%
- self-improving harness
- First native
Three pillars
Built for building — and evolving — agents
PenguinHarness is the first open-source harness to ship “agents building agents” and recursive self-improvement.
Simplest Is the Best
A deliberately minimal toolset over clean low-level interfaces: fewer tool calls, fewer tokens, complex tasks done efficiently.
Harness for Building Agents
With the Penguin SDK, an agent builds complete agent applications for you — autonomously, from scratch.
Harness for Recursive Self-Improvement
With Penguin Skills, an agent evaluates and optimizes itself, improving recursively over time.
vs. LangChain
Building agents with agents
With LangChain you build agents by hand — at 1× speed. With PenguinHarness, agents build agents — at 100×.
Agents built by hand
1×
LangChain
Chains, tools and prompts written line by line — every app starts from zero.
Agents built by agents
100×
PenguinHarness
One sentence in — an agent delivers scaffold, code and run instructions end to end.
Capabilities
The full capability set, one desktop-grade UI
Everything you used to install for — now online: multi-session chat, trace view, agent evaluation and more.
Multi-session chat
Any number of sessions per agent — streaming output, tool approvals and image paste out of the box.
Agent hub
Create and manage agents in one click; names, descriptions and prompts stay editable.
Skill library
Browse, install and quick-invoke skills — agents can write and optimize their own.
Trace view
Replay every request and tool call round by round, with token breakdown and timing.
Scheduled tasks
Cron-style schedules run agents on time, fully traced, unattended.
Agent evaluation
Built-in benchmark suites and scoreboards — scores keep climbing as agents evolve.
Built-in Skills
The built-in skill library at a glance
Four skill groups out of the box — agents can write and optimize their own, too.
Office Productivity
- data-analysis
- firecrawl
Software Development
- web-design
- software-engineering
AI App Development
- penguin-sdk
- penguin-cli
- agenthub-models
- vllm
- ollama
- llamafactory
Agent Tuning
- agent-creation
- benchmark-design
- agent-evaluation
- agent-optimization
Cases
From one sentence to a running app
Hand the requirement to an agent and get a runnable result end to end — right from the browser.
Collect the docs from https://github.com/ericbuess/claude-code-docs and build a RAG app that answers Claude Code questions as a configuration expert, citing its sources.
The generated RAG app: a Claude Code docs expert answering with cited, clickable sources and example questions.
And generating this entire RAG app burned just $0.02 of tokens — on DeepSeek V4 Pro.
Security
Evolution within bounds, data within walls
A runtime boundary designed for enterprise data security.
Open source, local deployment
A fully auditable open-source kernel; data lives in local directories and never passes through third-party services.
Bounded evolution
Self-improvement is strictly confined to workspaces and skills — the harness core security boundary is never modified.
Approvals & audit
Tool calls require user approval first, and every decision is written to the trace as an audit event.
Credential isolation
Credentials land as hidden 0600 files, are barred from the system prompt, and stay masked throughout the UI.
Quick start
Your first agent in three steps — no install
Everything runs online: open the console, type one sentence, and watch an agent build, run and improve your agent app.
Open the online console
Head to the Try page and open PenguinHarness right in your browser — nothing to download, nothing to configure.
Type one sentence
Describe the agent app you want, e.g. “Collect the docs from this repo and build a RAG app that answers questions with citations.”
Watch agents build agents
PenguinHarness scaffolds the code, runs the benchmarks, and ships version N+1 — every request observable in the trace view.
Community
Join the community and build with us
Discuss, ask, contribute — your first issue is the best way to start.
Discord
Chat with us and other developers in real time.
X (Twitter)
Follow the latest product and team updates.
WeChat group
Chinese community discussions and support.
GitHub
Stars, issues, and PRs all welcome.