Setting up an OpenSpace environment often stalls on version mismatches, missing credentials, and unclear steps for skill evolution. First verify you run Python 3.12 or newer; older runtimes cause import errors. If your Colab instance shows an older version, change the runtime type or create a local py312 venv before proceeding. Next obtain the required LLM keys—Anthropic or OpenAI—store them in environment variables, and set OPENSPACE_MODEL to your preferred model, e.g., anthropic/claude-sonnet-4-5.
Clone the repository with sparse checkout to avoid pulling large asset folders, then install the package in editable mode so changes are reflected instantly. Confirm the CLI tools openspace-mcp and openspace-dashboard are on your PATH. Create dedicated workspace and skill directories, write a .env file that exports OPENSPACE_MODEL, OPENSPACE_WORKSPACE, OPENSPACE_HOST_SKILL_DIRS, and any API keys, then source those variables into the shell.
When a key is present, launch asynchronous tasks via the OpenSpace Python API. The execute method returns a response plus any evolved skills, which are persisted in an SQLite database with versioning and lineage metadata. Inspect the database to see tables like skills, evolutions, and verify columns such as origin, name, and timestamps.
To encourage reuse, create a custom SKILL.md in your skill folder describing a simple capability—e.g., turning a CSV into a markdown report. Copy host skills like delegate-task and skill-discovery into the same directory so the agent can discover them. Run a second, related task; the system will attempt to FIX, DERIVE, or CAPTURE prior skills, lowering token cost on subsequent runs.
Start the MCP server with streamable‑http transport on localhost:8081, probe the endpoint to confirm it’s alive, and share the generated MCP configuration with external agents. If you have a cloud key, upload your custom skill to the OpenSpace registry; otherwise, keep it local. Finally, explore the showcase database to view skill origins and understand how evolution reduces cost over time.
This end‑to‑end flow solves the common pain points of environment setup, skill persistence, reusable agent behavior, and observable lineage, giving you a repeatable foundation for building self‑improving AI agents.
#AI #OpenSource #AgentEngineering #LLM #MCP #SkillEvolution