Coming from agentflow

10xGraph is the new name of Agentflow. The packages are now 10xgraph (core) and 10xgraph-api (server and CLI), imported as tenxgraph and tenxgraph_api, with the command 10xgraph and the config file 10xgraph.json. Old names still work until 2.0. This page walks through the rename.

What changed

Every old name has a new one, and the old one is a deprecated alias that is removed in 2.0. Use this table as a checklist for your project.

The graph model does not change. Nodes, edges, state, messages and the checkpointer API behave as before, so this is a rename and not a rewrite. Python 3.12 or later is required.

Replace the old packages with the new ones

Uninstall the old packages first, then install the new ones. Both generations ship a module called agentflow, so leaving the old packages in place causes conflicts.

# Remove the old distributions
pip uninstall 10xscale-agentflow 10xscale-agentflow-cli -y

# Install the new core and API packages (pick the provider extra you use)
pip install "10xgraph[google-genai]" 10xgraph-api

Other provider extras are openai and anthropic, and you can combine them, for example "10xgraph[google-genai,openai,anthropic]". Add storage extras such as pg_checkpoint or redis the same way if you use them. See Installation for the full list.

Update Python imports

Replace agentflow with tenxgraph and agentflow_cli with tenxgraph_api. The old imports keep working because they resolve to the same module objects as the new ones, but they emit a deprecation warning.

# Before
from agentflow import StateGraph, Agent, Message
from agentflow.storage.checkpointer import PgCheckpointer
from agentflow_cli import BaseAuth

# After
from tenxgraph import StateGraph, Agent, Message
from tenxgraph.storage.checkpointer import PgCheckpointer
from tenxgraph_api import BaseAuth

Only the top-level name changes. Submodule paths below it stay the same, so agentflow.core.graph becomes tenxgraph.core.graph. A search-and-replace of the two package names across your code and your 10xgraph.json agent path is enough. The reference lists the import path of every public symbol.

Rename the config file

The server reads 10xgraph.json first and falls back to agentflow.json when it is missing. Rename the file so there is no ambiguity. The keys are unchanged.

# Rename the project config (and any environment-specific variants)
mv agentflow.json 10xgraph.json

If you pass a path explicitly, use --config, for example 10xgraph api --config ./config/prod.json. The default value is 10xgraph.json. See Configure the server for every key.

Rename CLI environment variables

For each CLI variable the CLI checks TENXGRAPH_<NAME> first and uses AGENTFLOW_<NAME> only when the new one is not set. Rename any you set, on your machine, in CI and in container files.

Variables that configure your own app or the server, such as JWT_SECRET_KEY, REDIS_URL or DATABASE_URL, are not renamed. Core also still reads AGENTFLOW_LLM_TIMEOUT for the LLM client timeout.

Update the WebSocket subprotocol

Browser clients that authenticate a WebSocket with a bearer token send it in the Sec-WebSocket-Protocol header as two entries: the sentinel, then the token. The sentinel is now 10xgraph-bearer. The server still accepts the legacy agentflow-bearer sentinel.

// Before
// new WebSocket("ws://localhost:8000/v1/graph/ws", ["agentflow-bearer", token]);

// After: the sentinel first, then the JWT
const ws = new WebSocket("ws://localhost:8000/v1/graph/ws", ["10xgraph-bearer", token]);

The REST routes are unchanged. For the TypeScript client, see the client section.

Check the migration

Compile a graph with the new imports and start the server. This check does not call a model, so it needs no API key.

# check_migration.py
from tenxgraph import StateGraph
from tenxgraph.core.state import AgentState
from tenxgraph.storage.checkpointer import InMemoryCheckpointer
from tenxgraph.utils.constants import END


def hello(state: AgentState):
    return state


graph = StateGraph()
graph.add_node("HELLO", hello)
graph.add_edge("HELLO", END)
graph.set_entry_point("HELLO")

app = graph.compile(checkpointer=InMemoryCheckpointer())
print("Graph compiled with tenxgraph")

# Run the compile check, then start the API server with your renamed config
python check_migration.py
10xgraph api

Seeing no deprecation warnings means all imports are migrated. If the server starts from your renamed 10xgraph.json, the config rename worked.

Know what still works until 2.0

The old names keep working so you can migrate in stages. They are all removed in 2.0.

• import agentflow and import agentflow_cli resolve to the new modules and emit a deprecation warning.
• The agentflow command prints a deprecation notice and runs the same command as 10xgraph.
• agentflow.json is loaded when no 10xgraph.json exists.
• AGENTFLOW_* CLI variables are used when the TENXGRAPH_* one is not set.
• The agentflow-bearer WebSocket subprotocol is accepted.
• agentflow://media/ URIs and objects stored under the old media prefix still resolve. New media URIs are written as graph://media/.

Runtime paths and names

A few runtime locations moved. They are transparent in most projects, but know where they are if you script around them.

The SQLite checkpointer stores its default database at ~/.10xgraph/checkpointer.db. If ~/.10xgraph does not exist and ~/.agentflow does, it keeps using the old directory, so existing data is not orphaned. Setting an explicit path avoids the question.

Logger names changed from agentflow.* to tenxgraph.*. If you configure logging by name, for example logging.getLogger("agentflow.core"), change it to tenxgraph.core.

Fix common errors

ModuleNotFoundError: No module named ‘agentflow’

You uninstalled the old package but some code still imports from it. Change those imports to tenxgraph or tenxgraph_api.

The server cannot find its config file

10xgraph api looks for 10xgraph.json in the working directory. Run it from the project root, or pass --config with the path.

Deprecation warnings keep appearing

Something still imports agentflow or agentflow_cli, or you still run the agentflow command. Search your code, scripts, Dockerfiles and CI for the old names.

A WebSocket client fails the handshake

Check that the client offers exactly two protocol entries: 10xgraph-bearer (or the legacy agentflow-bearer) followed by the token. Browsers fail the handshake if the server does not confirm one of the offered subprotocols.

Next steps

• State graph: how graphs compile and run.
• Prebuilt agents: ready-made agent patterns.
• Configure the server: every 10xgraph.json key.
• Reference: import paths and signatures for the public API.