Getting Started¶
Install the core package when you only need discovery, configuration loading, lazy runner selection, local CLI runner profiles, and framework-neutral file tools:
pip install agentic-fabric
Install the runtime extras you actually use:
pip install "agentic-fabric[langgraph]"
pip install "agentic-fabric[strands]"
pip install "agentic-fabric[mcp]"
pip install "agentic-fabric[scraping]"
mcp exposes MCP integration dependencies. scraping installs the
first-party requests/BeautifulSoup scraping helpers. Vendor passthrough extras
are deferred until vendor-fabric is published with a stable optional-extra
contract. There is no aggregate AI or all-frameworks extra. Test, typing, docs,
and dev extras are reserved for repository validation and package maintenance.
Local CLI runners do not require a Python extra. They shell out to external executables that you install separately:
agentic-fabric list-runners --json
agentic-fabric run --runner aider --input "Add validation to auth.py"
Built-in filesystem tools also stay in the core package without depending on CrewAI or Pydantic. Runtime adapters add framework-specific wrapping only when their optional dependencies are installed.
from agentic_fabric import detect_framework, get_framework_info, get_runner
runtimes = get_framework_info()
framework = detect_framework()
runner = get_runner(framework)
Fabric agent definitions can live in .fabric/, .crewai/, .langgraph/,
or .strands/ directories. A .fabric/ directory is
framework-agnostic. A framework-specific directory requires that
runtime.
If a required runtime is unavailable, runtime errors include the matching install guidance. Automatic selection follows CrewAI, then LangGraph, then Strands, but explicit runtime selection always wins unless the fabric agent is stored in a framework-specific directory.
Use AgenticData when runtime context should travel with data and
registered fabric agent definitions:
from agentic_fabric import AgenticData
session = AgenticData({"repo": "jbcom/agentic-fabric"})
session.register_fabric_agent("reviewer", fabric_agent_config)
result = session.run_fabric_agent("reviewer", runtime="crewai")
CrewAI Dependency Note¶
The CrewAI adapter remains lazy and source-compatible, but
agentic-fabric does not publish a CrewAI extra while current CrewAI
releases depend on ChromaDB releases covered by an upstream critical advisory
with no patched version. Install CrewAI separately only after reviewing that
advisory state. Core, local CLI, LangGraph, Strands, MCP, and first-party
scraping installs do not include CrewAI.