Anything with an HTTP API, through the tool registry.
Why we built it
A workflow engine for the workthatcan't be wrong,and the moments thatneed a person.
Most automation tools are proud of how little they ask you. In finance and HR, that is the wrong instinct. Orkest is built so that a workflow which moves money or changes a record cannot be published at all until there is a human checkpoint in front of it — not a setting you remember to switch on, a rule the platform refuses to let you skip.
The platform
Boring where it counts
The interesting part of an automation platform should be your workflows. Everything underneath them should be predictable enough that you stop thinking about it.
Replay any run
Every node writes a row: what went in, what came out, how many tokens it burned and what it cost. Open a run from three weeks ago and the whole thing is still there.
extract_invoice2,904$0.0211
check_amount——
approval_1——
post_to_erp508$0.0073
Resume, don't restart
State is checkpointed to Postgres after every node. A worker can die mid-run and the retry picks up at the last completed step — it does not re-call the tool that already posted your invoice.
$0.0284
Median cost per run
Priced on what your workflows actually consume, at the node. Not per seat, and not per “task” with a definition that changes every quarter.
An audit trail nobody can quietly edit
Approvals, publishes, credential changes and quota rejections all land in one table — and the database itself refuses to change them afterwards. Not enforced by the application, where a bug or an admin could route around it.
CREATE TRIGGER audit_logs_no_update
BEFORE UPDATE OR DELETE ON audit_logs
FOR EACH ROW EXECUTE FUNCTION reject_audit_log_mutation();
Pricing
Free to start. Talk to us when it matters.
No seats, no credit card. You bring your own model key, so what the models cost is between you and OpenAI — we do not mark it up.
Start here
Free
$0to start
Unlimited workflows and workspaces
Up to 1,000 runs per day
Human approval gate on every mutating step
Immutable audit trail
Webhook, schedule and manual triggers
Knowledge base with cited retrieval
Bring your own OpenAI key — model usage bills to you at cost
For teams that need it running inside their own environment, or wired to systems we have not met yet.
Questions
The ones that come up
Something missing? The answer is probably specific to your stack —ask us directly.
It gets caught before it matters. Agent nodes return typed objects against a schema you declare, so a malformed answer fails the node rather than flowing downstream as plausible text. Anything that then writes to a real system sits behind an approval, and the approval screen shows you the upstream node's actual output as the evidence — not a summary of it.
Talk to us
Bring us your worst process
The month-end close nobody wants to own. The approval chain that lives in a spreadsheet. Describe it and we will tell you honestly whether Orkest is the right tool for it — and if it is not, what is.