Harness Profiles

Detailed comparison of each harness's strengths, limitations, and best use cases to help you choose the right one for your project.

🧠

Letta

openharness-letta

Strengths

  • Memory-first architecture
  • Full agent lifecycle (CRUD)
  • Memory blocks with semantic search
  • Archival memory persistence
  • Cloud API support

Limitations

  • No subagent delegation
  • Limited file operations
  • No hooks system
Persistent Memory Agent CRUD Semantic Search Cloud API
🦆

Goose

openharness-goose

Strengths

  • MCP-first architecture
  • Multi-model support (25+ providers)
  • Local CLI + Cloud REST API modes
  • Working directory context
  • Native skills via MCP extensions

Limitations

  • Stateless execution (no sessions)
  • Tools via config only
  • No persistent memory
MCP Native 25+ Providers Skills File Ops
⛓

Langchain Deep Agents

openharness-deepagent

Strengths

  • Built-in planning/todos
  • Native subagent delegation
  • Flexible filesystem backends
  • Multi-model via LangChain
  • File operations (read, write, glob, grep)

Limitations

  • No native skills API
  • Stateless by default
  • No hooks system
Planning Subagents File Ops LangGraph
💻

Claude Code

openharness-claude-code

Strengths

  • Full Claude Agent SDK integration
  • 12 built-in tools (Read, Write, Bash, etc.)
  • Native MCP server support
  • Skills system for reusable capabilities
  • Hooks for event-driven customization
  • Subagent spawning via Task tool

Limitations

  • No persistent memory
  • Sessions managed internally
  • Custom tools require MCP servers
Skills MCP Hooks Subagents Planning