About
It started with a report that had to be right every week.
SynapseOrch was started by Naveen Raj, an engineer who needed AI to write a weekly stock-market report: the same sources and the same sections, every week. The agent frameworks he tried chose their own route, and changing the model meant changing the code. So he built the engine he needed.
- Mon 22 Sep, 07:009 s · $0.0109
- Mon 29 Sep, 07:0010 s · $0.0112
- Mon 6 Oct, 07:009 s · $0.0111
The story
From one report to a product.
The first version had one job: a market-analysis workflow that produced the same report shape every week, from the same sources, without a model deciding halfway through that it would rather do something else. Every framework Naveen tried, CrewAI and LangGraph among them, made that hard, and made something else hard too: the model was baked in, so trying a better one meant touching the code.
What came out of it is an engine where the path is drawn before the run, every step names its model and can change it while a conversation is still going, the prompts and the memory are yours to edit, the documents an agent answers from sit in a knowledge base you can test, the bill stops at the plan’s limit, and a person signs off wherever it matters. It turned out a lot of businesses have the same problem, and most of them don’t want to own a framework to solve it.
It’s open source under AGPL-3.0, and it will stay that way. SynapseOrch Cloud is for teams who want it working today without running servers, not the “real” version.
What we believe
Four things the product is built on.
We believe
So the product
- Work should follow the path you approvedLetting a model choose its own next step is impressive in a demo and expensive in production. Most real work has a known shape, and the value is in running that shape reliably.
- So the product Draws the workflow as steps you can read, and shows every run on that path, step by step, with what each one cost. Change a step on the canvas and the next run uses it, with no code to touch.
- A person should sign off on what goes outAI is useful long before you’d let it send something on its own. The sign-off belongs in the workflow, where it can’t be skipped, not in a promise to check later.
- So the product Makes a sign-off a step: the run waits, for minutes or days, at no cost, and asks in the app or in Telegram.
- You should know the bill before it arrivesModel spend isn’t something to discover at the end of the month. A model chosen per step, branches that call no model, and a plan that stops at its limit all decide the bill in advance.
- So the product Lets you pick the model per step and decides branches and loops without calling a model; on Cloud it prices every step by its type and stops at your plan’s allowance instead of billing past it.
- Nobody should be able to take it awayModels change, and so do vendors. Switching models should be a dropdown, not a migration, and the engine under your workflows should be one you can run yourself.
- So the product Gives every agent and every model step a Model field you can change at any time, and is open source under AGPL-3.0: no license check, no telemetry, one line to install, and the same engine Cloud runs.
Where we are
Small, and not hiding it.
SynapseOrch is a small team. The open-source project is public and actively released, and its commit history is the honest record of what’s been built. SynapseOrch Cloud is open by invite, and the first 10 businesses get their first workflow built with us, free, because we’d rather get a few genuinely right first.
Pilots
No logo wall, because there’s nothing to show yet.
SynapseOrch Cloud is new. Rather than decorate this page with names we haven’t earned, here is who it suits today. Tick what sounds like you.
0of 8
Tick what sounds like you. These are the common ones, not the limit.
What you get
- A direct line to the people building it, not a support queue
- Your workflow set up with you, not handed to you as documentation
- Influence over what gets built next, while the roadmap is still soft
- Pricing agreed up front, and honesty about what is not ready yet
What we ask
- Real usage
- Honest feedback
- Permission to talk about the results, if they turn out well
Nothing gets published about your workflow without you agreeing to it first.
Request an inviteWorking here
How we work.
- The commit history is the interview
- Everything about the engine is public. If you want to know what working here looks like, read a few weeks of commits. That is the work itself, not a description of it.
- No open roles today
- We’d rather say that plainly than list positions we aren’t hiring for. When it changes, it’s posted here first, and the strongest introduction is a contribution to the open-source project.
A decision, as we write them
Light first
A first visit is light, whatever the visitor’s system says. Dark stays one click away and is remembered.
Considered and rejected: following the system setting; dark first.
Two ways to run it
Let us run it, with nothing to install and no API key needed. Or run it on your own servers, free and open source.
Free plan, no card needed.