agentdraft-langchain — LangChain tools for AgentDraft

LangChain tools that let an agent book calendars through AgentDraft — the coordination layer for AI scheduling agents — so two agents racing for the same slot resolve to a single deterministic winner instead of double-booking your calendar.

Each tool is a langchain_core.tools.BaseTool: LangChain and LangGraph consume it directly, CrewAI accepts LangChain tools natively, and AutoGen uses it through its LangChainToolAdapter.

Install

pip install agentdraft-langchain

Requires Python 3.9+, langchain-core>=0.3, and a valid avs_live_… agent key from the AgentDraft dashboard.

Quickstart

from agentdraft_langchain import get_agentdraft_tools

# Reads AGENTDRAFT_API_KEY (and optional AGENTDRAFT_BASE_URL) from the env.
tools = get_agentdraft_tools()

# Hand `tools` to any langchain-core-compatible agent. With LangGraph:
from langgraph.prebuilt import create_react_agent
from langchain_openai import ChatOpenAI

agent = create_react_agent(ChatOpenAI(model="gpt-4o"), tools)
agent.invoke({"messages": [("user", "Book me a 30-min call tomorrow at 2pm UTC.")]})

get_agentdraft_tools() returns two tools: agentdraft_get_availability and a booking tool. By default the booking tool is conflict-aware — see below.

The tools

All times are ISO 8601 strings (2026-06-01T14:00:00Z). The booking tools accept an optional idempotency_key for safe retries.

You can also construct tools directly against an explicit client:

from agentdraft import Client
from agentdraft_langchain import AvailabilityTool, ConflictAwareBookingTool

client = Client(api_key="avs_live_...")
tools = [AvailabilityTool(client=client), ConflictAwareBookingTool(client=client)]

How a lost race surfaces

When two agents race for the same slot, exactly one wins. The booking tools differ only in how the losing agent finds out:

• ConflictAwareBookingTool (default in get_agentdraft_tools()) returns a plain-language message so the LLM can react:

  OUTRANKED — your booking was not committed. A higher-priority agent
  (agent_id=agt_sales, priority=1) already holds this slot
  (winning_booking_id=bkg_…). Your priority is 3. Do not retry the same
  slot — call agentdraft_get_availability and propose a different time
  to your user.
  

  An autonomous agent reads this and proposes an alternate slot.

• BookingTool raises langchain_core.tools.ToolException with the same message. Use it when you want the surrounding framework to handle the failure explicitly.

A lost race is a normal, expected outcome of a coordination layer — not a bug. Prefer the conflict-aware tool for agents that run unattended.

Use with CrewAI / AutoGen

CrewAI accepts LangChain tools directly — pass the list get_agentdraft_tools() returns to a CrewAI Agent(tools=...). AutoGen wraps each tool with LangChainToolAdapter from autogen_ext.tools.langchain. Either way the conflict semantics are identical to the LangChain path.

How this differs from the other AgentDraft packages

All three wrap the same REST API with the same auth and semantics.

Links

• Protocol spec: https://agentdraft.io/spec
• API docs: https://agentdraft.io/docs
• Python SDK: agentdraft
• Changelog: https://agentdraft.io/changelog
• Source & issues: https://github.com/GipsyChef/agentdraft

Security

Found a vulnerability? See https://agentdraft.io/security — do not open a public issue for a security report.

Contributing

See CONTRIBUTING.md. This package lives inside the GipsyChef/agentdraft monorepo under sdks/langchain/.

License

MIT — see https://agentdraft.io/legal/license.