Agentflow is now 10xGraph, infrastructure for AI applications

Agentflow is now 10xGraph. Why we renamed it, what it is (self-hosted infrastructure for AI applications), what stays the same in your code, and the honest gaps.

Agentflow is now 10xGraph. The framework, the MIT license and the maintainers stay the same. The name changes because it was hard to find: many unrelated projects share the Agentflow name, and a search for it returns a Stanford research project, a Flowise feature and several other repositories before it returns ours. A name people cannot search for is a real cost for an open-source project, so we changed it.

What 10xGraph is

10xGraph is open-source, self-hosted infrastructure for AI applications. You write the agents, tools and tasks; 10xGraph runs them behind an API server with authentication, role-based access, tenant isolation, durable state in Postgres and Redis, background jobs, metrics and tracing, and generates Docker and Kubernetes deploy files. MIT licensed, by 10xScale. Formerly Agentflow.

Your agents are the core workload, and multi-agent handoff is one feature among many. The point is the infrastructure around them: the part every team ends up building before an AI feature can ship to its own customers. The tagline is “10xGraph by 10xScale: graph engineering for production AI agents.” The product promise is shorter: write your AI logic, and 10xGraph runs the infrastructure.

What it runs for you

  1. Security and access on every endpoint. JWT or custom auth, roles with a reach of own, tenant or all, and ownership enforced in storage: a thread outside the caller’s reach answers 404. The server refuses to boot if a route has no permission guard. Auth is off until you configure it; the production template turns it on. See Auth and the platform overview.
  2. Data and state. Threads live in Postgres with Redis as a hot cache, and writes are versioned so two runs on one thread cannot overwrite each other. Memory and RAG can use pgvector in the same Postgres. Tenant isolation uses columns you declare, filled from the verified identity. See Multi-tenant data and Checkpointing and threads.
  3. Background jobs. A thin TaskIQ layer: 10xgraph worker, 10xgraph scheduler and /v1/tasks routes. Delivery is at-least-once, so write tasks that are safe to rerun. See Background tasks.
  4. Observability. Tracing through OTLP, Phoenix, Logfire or LangSmith, Prometheus or OTLP metrics for the API and the worker, JSON logs, and a metadata-only audit log that you ship to your own storage. See Observability and Audit log. For teams going through SOC 2 and ISO 27001 audits, the control mapping says which controls the platform supports and which stay with you.
  5. Correct under failure. Finished tool calls are recorded and not run again when a run resumes after a crash, and nodes and tools have real timeouts. The limit: a crash before a tool returns can repeat it, so send idempotency keys to anything that moves money. Read Replay-safe tools and Your agent charged the card twice.
  6. Deploy it yourself. 10xgraph build writes a Dockerfile, a Compose file and a Kubernetes manifest. See Deploy.
  7. One stack, backend to frontend. A typed TypeScript client covers invoke, stream, threads, memory and files, and tools can run in the user’s browser. See Connecting clients and Remote tools.
  8. You own it. MIT licensed, self-hosted, no hosted service to depend on, and not tied to one model vendor: OpenAI, Google Gemini, Anthropic and any OpenAI-compatible endpoint. See Providers.

Graph orchestration, tool-calling loops, MCP and streaming are in there too, but every serious framework has those. The list above is why you would pick this one.

Since then: the hardening that followed the rename is written up in What production-ready means in 10xGraph, and the full list of defaults is on the platform overview.

What stays the same

  • The project and the people. Same code base, same maintainers, same license: MIT, made by 10xScale.
  • Configuration keys. The keys in the configuration file stay the same. Only the file name changes, from agentflow.json to 10xgraph.json, and agentflow.json is still read until 2.0.
  • Your server setup keeps running. In 10xgraph-api the agentflow command, from agentflow_cli import BaseAuth and the AGENTFLOW_* CLI variables still work as deprecated aliases until 2.0.
  • Your existing code keeps running. from agentflow import StateGraph still works: agentflow stays importable as a deprecated alias until 2.0. New code should import from tenxgraph, as in from tenxgraph import StateGraph.

What changes

Before After
pip install 10xscale-agentflow pip install 10xgraph
from agentflow import StateGraph from tenxgraph import StateGraph
pip install 10xscale-agentflow-cli pip install 10xgraph-api
npm install @10xscale/agentflow-client npm install 10xgraph-client
CLI command agentflow CLI command 10xgraph
Config file agentflow.json Config file 10xgraph.json
GitHub organization 10xHub GitHub organization 10xGraph

The final releases under the old names are 10xscale-agentflow 0.10.1, 10xscale-agentflow-cli 0.6.1 and @10xscale/agentflow-client 0.5.0. The first releases under the new names are 10xgraph 0.10.0 and 10xgraph-api 0.7.0 on PyPI, and 10xgraph-client 0.6.0 on npm. New releases and fixes go to the new package names.

Honest weak spots

You should know where we are behind before you choose us.

  • A smaller community. There are fewer integrations, tutorials and answered questions than for LangGraph or CrewAI.
  • The rename resets recognition. 10xGraph has no search history yet, so existing blog posts and answers still say Agentflow.
  • Visual tooling is lighter. LangGraph has a stronger visual studio and hosted tracing. We have a React playground and OpenTelemetry, Phoenix, Logfire and LangSmith tracing.
  • No hosted service. You run it, or nobody does.
  • Auth is off until configured, and tenant isolation needs columns you declare. Neither is automatic.
  • Python 3.12 or newer, and code-first. This is not a no-code builder.

The roadmap lists what is missing or partial.

Next steps

Written by

Shudipto Trafder

Maintainer, 10xGraph

Keep reading

Subscribe via RSS

5 min read

A prompt-injected model cannot skip your approval step

How 10xGraph keeps handoffs, tool visibility, tool results and interrupt resumes under the graph's control, so a steered model cannot walk around a human approval.

  • security
  • interrupts
  • human-in-the-loop

Blog home Archive