LangChain, LangGraph and Deep Agents
Not in the preview yet. This feature is built in the repository but is off in the v2.2.6 preview (test mode). Do not treat this page as available on a preview core. See ../preview.md and the release notes.
dynamoprotocol-langchain adds Dynamo to LangChain 1.x, LangGraph and Deep Agents. One plugin covers all three. Every agent in a run gets its own capped slice of one budget. A LangGraph hand-off or a Deep Agents sub-agent opens a child slice, which can never exceed what its parent has left. A model call that would go over its slice is refused before the model is called.
The plugin is not published. It installs from a clone of the repository.
Install and try it
These are the commands from the plugin README. They boot a local core in Control Mode, build a venv from the hash-checked locks/test.txt and run the offline containment demo on a scripted mock chat model: no provider key, no model called. You need Node 22.12 or later, pnpm 9, curl, openssl and Python 3.11 (set PYTHON=python3.11 if python3 is another version). Set DYNAMO_REPO to your clone.
cd "$DYNAMO_REPO"
pwd
pnpm install --frozen-lockfile
pnpm --filter @dynamo/core-host... -r build
export DYNAMO_OWNER_TOKEN="dp-local-owner-$(openssl rand -hex 16)"
export DYNAMO_CORE_URL="http://127.0.0.1:${DYNAMO_CORE_PORT:-8790}"
[ -s .dynamo-local.key ] || (umask 077 && openssl rand -hex 32 > .dynamo-local.key)
export CORE_KEY_ENCRYPTION_KEY="$(cat .dynamo-local.key)"
OWNER_API_TOKEN="$DYNAMO_OWNER_TOKEN" PLUGIN_WIRE_TOKEN="dp-local-wire-$(openssl rand -hex 16)" \
OWNER_API_PORT="${DYNAMO_CORE_PORT:-8790}" CORE_DB="$PWD/.dynamo-local/core" CORE_LOG=off \
CORE_RATE_LIMIT_RPS=200 CORE_RATE_LIMIT_BURST=400 \
node deploy/image/entry.mjs > .dynamo-local.log 2>&1 &
CORE_PID=$!
trap 'kill "$CORE_PID"' EXIT
"${PYTHON:-python3}" -m venv .venv-langchain
PIP_ONLY_BINARY=":all:" .venv-langchain/bin/pip install --quiet --require-hashes -r packages/plugin-langchain/locks/test.txt
.venv-langchain/bin/pip install --quiet --no-deps ./packages/sdk-python ./packages/plugin-langchain
for _ in $(seq 150); do curl -fsS "$DYNAMO_CORE_URL/healthz" > /dev/null 2>&1 && break; sleep 0.2; done
kill -0 "$CORE_PID" 2>/dev/null && curl -fsS "$DYNAMO_CORE_URL/healthz" > /dev/null || tail -n 20 .dynamo-local.log
.venv-langchain/bin/python packages/plugin-langchain/examples/containment/run.py
kill "$CORE_PID"; trap - EXIT
The last line stops the core. Delete .venv-langchain when you are done.
Wrap an agent run in a budget
This is packages/plugin-langchain/examples/app/with_dynamo.py: a small three-agent LangGraph app whose agents hand off to each other, with Dynamo added. make test-plugin-langchain runs it against a local core (tests/test_examples.py).
# SPDX-License-Identifier: Apache-2.0
"""A small multi-agent app on LangChain + LangGraph — with Dynamo.
Runs offline: the scripted mock chat model stands in for a real one.
"""
import asyncio
from typing import Annotated
from langchain.agents import create_agent
from langchain_core.messages import HumanMessage, ToolMessage
from langchain_core.tools import InjectedToolCallId, tool
from langgraph.graph import START, MessagesState, StateGraph
from langgraph.types import Command
from dynamoprotocol_langchain import AgentConfig, DynamoLangChain, usd
from dynamoprotocol.testing import mock_prices
from dynamoprotocol_langchain.testing import ScriptedChatModel, Text, ToolCall, Upstream
upstream = Upstream()
SCRIPT = {"orchestrator": [ToolCall("transfer_to_researcher")], "researcher": [Text("three sources found")], "writer": [Text("draft written")]}
def model(name):
m = ScriptedChatModel(model_name="mock/scripted-1", upstream=upstream, label=name, max_tokens=256)
m.queue.extend(SCRIPT[name])
return m
def handoff(agent):
@tool(f"transfer_to_{agent}")
def transfer(tool_call_id: Annotated[str, InjectedToolCallId]) -> Command:
"""Hand the conversation to another agent."""
msg = ToolMessage(f"transferred to {agent}", tool_call_id=tool_call_id)
return Command(goto=agent, graph=Command.PARENT, update={"messages": [msg]})
return transfer
async def main():
dyn = await DynamoLangChain.connect(
prices=mock_prices(), cap=usd("5"), mandate={"categories": ["research", "writing", "review"], "maxDepth": 3},
agents={"orchestrator": AgentConfig(usd("5"), "review"), "researcher": AgentConfig(usd("1"), "research"), "writer": AgentConfig(usd("1"), "writing")},
)
builder = StateGraph(MessagesState)
for name in SCRIPT:
others = [o for o in SCRIPT if o != name]
builder.add_node(name, create_agent(model(name), [dyn.handoff_tool(o) for o in others], middleware=[dyn.middleware(name)], name=name), destinations=tuple(others))
builder.add_edge(START, "orchestrator")
graph = builder.compile(name="orchestrator")
result = await dyn.run(graph, {"messages": [HumanMessage("Write a brief on grid-scale storage")]})
print(result["messages"][-1].content)
await dyn.close()
asyncio.run(main())
DynamoLangChain.connect(...)opens a budget with your cap and mandate. It
reads DYNAMO_CORE_URL and DYNAMO_OWNER_TOKEN when you do not pass core_url and token.
dyn.middleware(name)goes last in the agent's middleware list. It
reserves the call's worst case before the model is called, and bills the usage the model reported after it.
dyn.handoff_tool(agent)opens the target's child slice before the graph
routes to it.
dyn.run(graph, input)replacesgraph.ainvoke(input).
With a real model, pass your price table instead of mock_prices() (for example prices="quickstart/dynamo-guard/prices.json") or set DYNAMO_PRICE_TABLE. A model with no row is refused.
What a refusal looks like
A refusal is a LangChain exception (a LangChainException subclass) that is also one of the layer's DynamoAgentError classes, so you can catch either: DynamoCapExceeded, DynamoMandateRefused, DynamoRevoked, DynamoUnpricedModel, DynamoFrozen, DynamoStopped, DynamoApprovalPending, DynamoApprovalDenied and DynamoAmendmentRefused. Each carries stream_id, reason and headroom. The refused call never reaches the model.
A model without a completion limit (max_tokens, max_completion_tokens or max_output_tokens) and without default_completion_tokens is refused rather than guessed. A Deep Agents sub-agent without a Dynamo middleware is refused before it runs.
Funding modes
- Control Mode (Live). A budget the plugin opens itself (
cap=) is
always a Control Mode budget: every check runs and no money moves. The quickstart and every test of this plugin use one.
- A funded budget. You can pass an existing stablecoin or credit card
budget by id (budget_id=). The plugin does not refuse it, but no test runs the plugin on one. In those modes the core requires signed hops by default, and a child slice opened without its delegator's signature is refused hop_unsigned. Pass hops= (a LayerHops from dynamoprotocol.agents: the delegator's key and each payee's own signed offer) so the plugin signs every hop it opens.
- Real money (Release-gated). A core runs a money mode only when its
operator allows it, and real-money settlement is not released.
Known limits
Read packages/plugin-langchain/KNOWN-LIMITS.md in the repository before you rely on the plugin.