API reference
The full public API exported from langstitch. Import anything from the top level:
from langstitch import (
graph_node, graph, skill,
input_guardrail, output_guardrail,
business_policy, persona, configuration,
langstitch_graph_server, GraphBuilder, LangStitchApp,
load_config, get_registry, START, END,
)
Decorators
| Symbol |
Kind |
Purpose |
@graph_node |
decorator |
node handler state → dict |
@graph |
decorator |
graph builder (entrypoint=, parent=) |
@skill |
decorator |
reusable capability |
@input_guardrail / @output_guardrail |
decorator |
inbound/outbound validation |
@business_policy |
decorator |
priority-ordered org rule |
@persona |
decorator |
identity / system prompt |
@configuration |
decorator |
bind a config section to a dataclass |
@tool / @worker_agent |
decorator |
LLM tool / delegatable sub-agent |
@langstitch_graph_server |
decorator |
graph API server (protocol, port, name, properties) |
MCP
| Symbol |
Kind |
Purpose |
@langstitch_mcp_server |
decorator |
MCP server class + transport |
@mcp_tool |
decorator |
expose a callable as an MCP tool |
@mcp_resource |
decorator |
expose a readable MCP resource |
@mcp_prompt |
decorator |
expose a reusable MCP prompt |
Graph
| Symbol |
Kind |
Purpose |
GraphBuilder |
class |
declarative wiring → LangGraph |
START, END |
const |
edge sentinels |
Runtime
| Symbol |
Kind |
Purpose |
LangStitchApp |
class |
runtime: bootstrap(), build_graph(), invoke(), info() |
create_app |
fn |
build the FastAPI app (needs server) |
run |
fn |
start the server |
Providers & helpers
| Symbol |
Kind |
Purpose |
get_config, reset_config_cache |
fn |
cached config access |
get_env, get_secret |
fn |
environment & secret access |
get_logger |
fn |
configured logger |
get_llm_provider |
fn |
chat model from model: config |
get_http_client, get_async_http_client |
fn |
configured httpx clients |
External services
| Symbol |
Kind |
Purpose |
AuthConfig, ServiceConfig |
dataclass |
service definitions |
ServiceClient, AsyncServiceClient |
class |
typed HTTP clients |
load_service_config, format_path |
fn |
service helpers |
set_request_headers, get_request_headers, clear_request_headers |
fn |
header propagation |
Config
| Symbol |
Kind |
Purpose |
AppConfig |
class |
parsed configuration object |
load_config, load_application_config, load_env |
fn |
load config & env |
find_project_root |
fn |
locate the project root |
query_json |
fn |
JSON-path lookups |
compile_config |
fn |
YAML → JSON precompile |
Registry
| Symbol |
Kind |
Purpose |
Registry, get_registry, reset_registry |
class/fn |
the process-global registry |
Dynamic registries
| Symbol |
Kind |
Purpose |
ToolRegistry, AgentRegistry |
class |
lazy spec registries |
get_tool_registry, get_agent_registry |
fn |
registry accessors |
get_all_tools, get_all_worker_agents |
fn |
list specs |
get_input_guardrails, get_output_guardrails |
fn |
list guardrail specs |
get_skills, get_policies, get_personas |
fn |
list specs |
refresh_registries |
fn |
force a refresh |
Hierarchical context
| Symbol |
Kind |
Purpose |
Context, ContextScope, ContextBuilder |
class |
context primitives |
LLMContext |
class |
the object a model callback receives |
run_llm, run_worker_agent |
fn |
run an isolated LLM / sub-agent call |
Source of truth
Exhaustive signatures and options ship with the langstitch-sdk package on
PyPI — inspect them via
python -c "help('langstitch')" after install.