Business-grade AI agents
Every agent is a workflow: a plan you can see, change and test. Improve it using your real runs, then put it in production on an engine that retries from the failed step and traces everything it does. Fair Source, on your own infrastructure.
Why workflows
Ask an AI agent to do a job and it works out the steps on the spot. Ask again and it may choose different ones. You cannot test that, audit it or promise it. A workflow is the same job with the steps written down, so you can. AI is used only where a step needs it, and every step is recorded.
Decide the steps once.
A free-form agent works out its steps again on every run, so results vary and nobody can say why. In Faraflow you lay the steps out on a visual canvas and save them as a versioned workflow that your team reviews and owns. You know what it will do before it does it.
Built for what happens after launch.
Real systems time out and inputs arrive messy, and demos do not get retries or traces. Faraflow’s engine does: it retries from the failed step, traces every step, and runs on your infrastructure. When asked “why?”, you can point to the step that did it. With predictable bills.
Get good fast on real data.
AI makes mistakes, and an agent is only as good as your ability to iterate on it. Faraflow keeps each pass small: change a step, test it on saved cases, compare with the last run. Load any past run to see which step went wrong, then fix that step. Small, focused AI steps with the right context are easier to get right.
| Aspect | Free-form agent | Faraflow workflow |
|---|---|---|
| When a run fails | Run the whole thing again. | Retry from the step that failed. |
| When asked “why?” | Reread a long chat and guess. | Point to the step and data behind it. |
| Changing behavior | Edit the prompt and hope. | Change one step and test it before it goes live. |
| What you own | A chat history. | A versioned plan and the test cases that prove it. |
| Choosing models | One model for the whole job. | The right model for each step, from 100+ providers. |
| Using it elsewhere | Lives in a chat or script. | Publish it as an API. |
Run the whole thing again.
Retry from the step that failed.
Reread a long chat and guess.
Point to the step and data behind it.
Edit the prompt and hope.
Change one step and test it before it goes live.
A chat history.
A versioned plan and the test cases that prove it.
One model for the whole job.
The right model for each step, from 100+ providers.
Lives in a chat or script.
Publish it as an API.
Capabilities · Live today
Everything here works in Faraflow now. Anything still to come is on the roadmap, marked that way.
DriveUp is an AI visibility analysis tool for self-storage operators integrated with multiple LLM providers and many other sources. Its report workflows run on the Faraflow engine, in production.
selfstorageapps.com
Every node shows its latest outputs right on the canvas.
AI, code, logic and integration nodes, plus your own custom nodes.
See each run node by node, and load any past run back onto the canvas.
Step through a node’s outputs, config and connections, upstream and downstream.
Keep real inputs as test cases and rerun the workflow against them.
Fair Source, so you can see and run everything.
Backed by Alokin, so you’re never on your own.
Faraflow is Fair Source. The code is public, free to inspect, and free to run on your own infrastructure for your own use. We use the Fair Core License (FCL-Apache 2.0) to stop anyone, cloud providers included, from selling our engine as a service, so we can keep funding the project. And every release automatically becomes fully open source under Apache 2.0 after two years.
The big advantage: nothing in Faraflow is locked behind a license key today. Enterprise features will live in the same public codebase as everything else, so anyone can read exactly what a key unlocks before they commit. Only those features will need a key to run; the majority of Faraflow remains available to everyone without one.
The code is public. Inspect what your workflows run on, and change it if you need to.
On your own infrastructure, for your own use. Consultants and agencies can deploy it for their clients and sell professional services around it.
Every release becomes fully open source under Apache 2.0 after two years, so you are never dependent on a single vendor.
Alokin supports Faraflow deployments for teams that need someone to call.
Alokin’s engineers can design and build workflows for your business and connect them to your systems.
Alokin maintains and extends Faraflow; the roadmap below shows what’s next.
Developed and commercially supported by Alokin, a software consultancy delivering software since 2003, including SaaS, web/mobile development, APIs, AI/NLP, e-governance and enterprise systems.
Faraflow grew out of the open-source PySpur project and has been heavily extended since.
Roadmap
We’re constantly building and improving Faraflow. Here’s what’s coming next.
Your use case can reshape this list.
Tell us what matters to you.
FAQ
What people ask us most about Faraflow. Anything else, ask us directly below.
Faraflow is a Fair Source platform for building AI agents and workflows, developed and supported by Alokin. You design a workflow visually in the studio, run it with the engine on your own infrastructure, and trace every step of every run.
Faraflow is Fair Source, a deliberate choice: the code is public, free to inspect, and free to run on your own infrastructure for your own use. No current feature needs a license key; enterprise features, when they arrive, will. It is licensed under the Fair Core License (FCL-Apache 2.0), which stops anyone offering Faraflow itself as a competing product or service. We want Faraflow to be sustainable to build and support. Every release becomes fully open source under Apache 2.0 two years after it is published. Faraflow grew out of the open-source PySpur project and has been heavily extended since. View the code on GitHub
Yes. Consultants and agencies can sell professional services and deploy Faraflow for their clients to run for their own use. You can also build your own products on top of it. What needs an agreement with Alokin is offering Faraflow itself, or a service that substitutes for it, to other customers as your own product.
Yes. Nothing in Faraflow is locked behind a license key today. Enterprise features will live in the same public codebase as the rest of Faraflow, so anyone can read the code and see exactly what a license key unlocks. Only those features will need a key to run, which the Fair Core License protects; the majority of Faraflow remains available to everyone without one. That is the advantage of the Fair Core License: nothing has to be held back in a closed repository.
Yes. Faraflow runs on your own infrastructure, so you decide where your workflows run and where your data lives.
An agent plans its work again on every run. In Faraflow the steps are decided once, a model is called only where a step needs one, and every run follows the same path. You can still use a free-form agent inside a step where the task is open-ended.
LangChain and LangGraph are code libraries: they give developers building blocks for writing agents and graphs in Python or JavaScript. Faraflow is a complete studio and engine. You design the workflow visually, run it on your own infrastructure, and get run traces, test cases, retries and publishing as an API built in, so the whole team can see and change how a workflow works. Code nodes let developers add custom logic where they need it.
Faraflow connects to 100+ model providers, and you choose the model per step: a smaller, cheaper model for routine steps and a larger one where judgment matters.
Yes. DriveUp, an AI visibility analysis tool for self-storage operators, runs its report workflows on the Faraflow engine in production. There are no other named customers yet, and we use Faraflow ourselves.
Alokin, a software consultancy delivering software since 2003. Alokin maintains Faraflow, offers commercial support, and can design and build workflows for your business.
Any digital workflow with a clear sequence, for example analysing a student’s performance from a questionnaire. Today you can design workflows visually, run them on your own infrastructure, choose among 100+ models per step, trace and retry runs, test steps on saved cases, and publish a workflow as an API. Voice, browser agents, SMS and WhatsApp, and cost tracking are on the roadmap and not available yet. See use-case ideas
Faraflow works with 100+ model providers, plus AI, code, logic and integration nodes, and custom nodes for your own tools. Integrations today include Slack, GitHub, Google Sheets, Firecrawl, Exa, Reddit, Jina, YouTube transcripts and Email. Founder Partners can influence which integrations are built first. A number of integrations are on the roadmap.
Tell us what you are trying to solve, and we will show you how it could work as a workflow.
Meet us in person
Come by for a live walkthrough of the studio and the engine, or bring the workflow you have in mind.
Prefer email? rajeev@alokin.in