Plans and Credits

SynapseOrch Cloud has two costs, and they are separate on purpose.

Counts are not priced. Agents and workflows cost nothing to hold. Their ceilings appear in the table below because they describe scale, not because they drive the bill. Two workflows firing every minute cost far more to serve than twenty that run monthly, so execution is what is charged.

The plans

Paid yearly, each plan costs ten times its monthly price. For more than Business offers, a Custom plan is agreed with us: get in touch.

Every plan stops at its credits rather than billing you. Warnings arrive when 50%, 25% and 10% of the month's credits are left. When they run out, runs stop until the next period: buy more credits under Settings → Billing (they last twelve months), or move to a bigger plan. Nothing is ever billed past what you chose.

A run that is already working may finish a small shortfall — up to 5% of the month's credits, 50 on Free — so a run you are watching isn't cut a step from its end. A larger one stops before the step it can't pay for, and says so: Stopped: this workspace is out of execution credits. This run needed about 120 credits and 40 were left for it. Add credits under Settings, Billing to continue.

Single sign-on is coming and is not available yet, on any plan.

Vault storage counts every file in the Vault, a knowledge base's documents and fetched pages included; the index Knowledge builds from them does not count.

Limits by plan

The other things each plan sets. Counts like these describe scale; they are not what you pay for.

At a limit, whatever would pass it is refused, and says what to do instead.

What a credit is

A step is priced by its type:

A run costs its steps' prices added up — each step inside a Loop or a Parallel counted as it runs — or one credit per started minute of active time, whichever is more, and at least one credit. For ordinary workflows the steps decide it, because they finish in seconds.

Some work is priced its own way:

Model tokens are billed apart from credits, as they are used: from the model wallet, or nothing at all on your own keys.

Max run length

Max run length is a runaway guard, not an allowance. It stops a broken loop consuming a plan; it is not a budget you are expected to use.

• It counts working time. Waiting for a person, or in line for a slot, does not count.
• It is set when the run starts, from the plan the workspace has then, so a plan change mid-run neither cuts it short nor lets it run on.
• The tighter limit holds: the plan's length or the workflow's own timeout, whichever is shorter. A step in progress is cut at what is left.
• A run past it ends failed, and says why: Stopped at the 30-minute run limit of this workspace's plan, after 30.1 min of work — time waiting for a person is not counted. It is charged for what ran, the step that was cut included.

Models

Platform models are available on every plan, including Free, funded by a prepaid wallet rather than gated by tier. See Models and the wallet.

Bring your own provider keys instead and the model half of your bill is zero — you pay your provider directly and SynapseOrch Cloud only charges for execution. Both approaches work at once: platform models for some agents, your own keys for others.

Where a model is chosen automatically, the model actually used is recorded in the run trace, so what you were charged for is always checkable.

Builders and viewers

A builder is anyone who can create or edit agents, workflows and tools: builders, admins and owners. Viewers, who read workflows, runs and results, and approvers, who can also answer what waits for a person, are free on every plan and never count toward its builders.

Each paid plan includes some builder seats, then adds builders at its per-seat rate up to its ceiling (the table's Builders row); the charge is shown and confirmed before it is applied. Free sells no extra seats: two people can build, and everyone else is a viewer. Pending invitations count until they are accepted or revoked. See Organisations, roles and seats.

Capacity

Runs at once is how many of a workspace's runs execute at the same time. A run started beyond it waits in line, first in, first out, and starts the moment one of the workspace's runs finishes. Nothing is dropped, and a scheduled run waits too. The run's page says its place (Queued, 2nd in line) and moves up on its own. Synapse AI waits in line like any run, and says Waiting for one of this workspace's runs to finish (2nd in line)….

• Waiting costs nothing. No credits are held or charged while a run waits; they are held when it starts. A workspace without the credits to start the run even now is told so at once. If its credits run out while the run waits, the run ends at its turn, uncharged: Not started: when its turn in the queue came, this workspace did not have enough execution credits for it. Nothing was charged.
• Cancel a waiting run and it ends at once, at no cost.
• The line has a length: ten times the plan's runs at once (the table's Can wait in line). Past it, a new run is refused, and a schedule skips that occurrence and says so.
• Some things never wait: a person's answer to a paused run, and a Replay. Knowledge evaluations don't wait in line either: as many run at once as the plan's runs at once, counted apart from workflow runs, and a new one is refused while that many are under way.

One workspace's runs never go ahead of another's: there is no priority between plans.

Fair use

SynapseOrch Cloud does not sell "unlimited" anything. Limits that are described as high and fair-use have a named ceiling, and that ceiling is raised on request rather than enforced silently.

Getting a plan

Sign up and create an organisation, then choose a plan under Settings → Plan.

Self-hosting is free and unlimited under AGPL-3.0, and always will be. See Cloud vs self-host.