SW·003.1 - Automations

Draw the process.
Run it forever.

Automations are built on a canvas, not in a config file. Drop in a trigger, wire it to an assistant, branch on a condition, pause at a named person, call Xero. What you see is what runs, and every run leaves a receipt.

See it running
order_to_cash · builder
RUN
ADD NODE Trigger Assistant Condition Human gate Tool Loop Merge Wait Sub-workflow
Schematic of the automation builder: a trigger, an assistant, a condition that branches to a human gate or straight to Xero, a merge, and a Slack notification yes no TRIGGER quote_accepted ASSISTANT ops_assistant CONDITION amount > 5k? HUMAN GATE ⏸ finance review TOOL xero.invoice.new MERGE join branches TOOL slack #sales
7 nodes · 1 human gate · 2 integrations validation: ready to run
SCHEMATIC - the builder canvas. Screenshot slot: app/assets/images/synth_wise/screen_shots/landing/fig-03-builder-canvas.png

Visual building

Simple enough to see.
Specific enough to ship.

The canvas is the specification. Nodes are typed, connections are checked as you draw, and the run follows the picture.

01 - CANVAS

Draw it, then run it

Nodes on a canvas, edges between them, a settings panel for the node you have selected. Validation badges show what is missing before you press run, not after.

typed nodes · live validation · versions

02 - TRIGGERS

Start from anything

A webhook from Quoter, a schedule, a chat message, a new report record, a polled mailbox, a button in the dashboard. One automation can start another.

webhook · schedule · chat · event · poll · manual

03 - NODES

Branch, loop, merge, wait

Conditions route the run. Loops walk a list. Merge brings branches back together. Wait holds for a time or a signal. Sub-workflows keep the big picture readable.

condition · loop · merge · wait · sub-workflow

04 - HUMAN GATES

A person is a node

The run pauses, a named person sees the item for review, approves or edits it, and the run resumes. Who decided, and when, is persisted with the execution.

review items · approval inputs · audit trail

05 - INTEGRATIONS

Your systems, as tools

Xero, Slack, HubSpot, Trello and Google Calendar out of the box. Anything with a REST or webhook API joins through the marketplace. Credentials stay encrypted in your workspace.

marketplace · REST · webhooks · encrypted credentials

06 - DATA

Map, look up, write back

Map fields between nodes without code. Pull reference data from lookup tables your team maintains. Write distilled documents back into a context library for the next run.

data mapping · lookup tables · context library writes

AI in the flow

Agents inside.
Bounded by design.

An assistant node is a role with instructions, tools and project access. In a flow it acts as an agent, taking the multi-step action the run needs, inside the guardrails you set.

01

You choose the model and the bounds

Pick the model per node, frontier or self-hosted. The assistant only sees the context and tools its project allows. Workspaces ring-fence what each run may touch.

02

More than one lookup

Summarise a document set, chase every overdue supplier, draft the reply and raise the ticket. Each tool call is a visible step, not a black box.

03

Every call receipted

Model, provider, tokens, latency, who triggered it, which execution and which node. The inference log is written as the run happens.

AGENT SESSION
ops_assistant · tools: xero, slack, trello
Chase every supplier with an overdue PO.
  1. xero.purchase_orders.list status=overdue days>5 3 results
  2. slack.post #suppliers · per PO 3 sent
  3. trello.card.create list=Follow-ups 3 cards
  4. provenance.record 3 model calls · 2,614 tokens INTERNAL
acting inside the assistant's permissions every call receipted

Debug and monitor

See the run.
Fix the step.

Executions are first-class records. Open one and read what each node received, produced and decided.

MONITOR

Executions on the dashboard

  • Pin an automation and watch its runs from the workspace dashboard
  • Status, duration and the node a run is waiting on
  • An executions summary per automation: runs, failures, last run

DIAGNOSE

Node-level logs

  • Input and output for every node in the execution
  • The inference log beside it: model, tokens, cost
  • Validation badges flag the misconfigured node before it runs

FIX

Edit without breaking live runs

  • Versions: edit the draft while the published version keeps running
  • Connection stats show what flowed between two nodes
  • Import and export definitions as JSON

TEST

Run it by hand

  • Manual triggers with a form for the inputs
  • Chat triggers: start it from a conversation
  • Open the execution while it runs; steps appear as they finish
AUTOMATION RUN
order_to_cash · run #4127
# step system state
  1. 01 quote_accepted Quoter DONE
  2. 02 contact → project → invoice Xero DONE
  3. 03 ⏸ payment gate human APPROVED
  4. 04 po_split_by_manufacturer Xero RUNNING
  5. 05 supplier_dispatch_watch supplier API QUEUED
  6. 06 carrier_booking carrier QUEUED
  7. 07 proof_of_delivery → final Xero QUEUED
4 integrations · 1 human gate a receipt on every step
FIG·04 - an execution opened to node level, inputs and outputs expanded. Place the screenshot at app/assets/images/synth_wise/screen_shots/landing/fig-04-execution-panel.png

The builder, as it is

Not a mock-up.
The product.

The drawings on this page are schematics. These frames hold screenshots of the live builder and a paused human gate. Until they land, the slot says so rather than pretending.

FIG·03 - the builder canvas with a node selected and its settings panel open. Place the screenshot at app/assets/images/synth_wise/screen_shots/landing/fig-03-builder-canvas.png
FIG·05 - a run paused at a human gate, review item and approve action visible. Place the screenshot at app/assets/images/synth_wise/screen_shots/landing/fig-05-human-gate.png

Deploy and self-host

Yours to run.
Yours to leave.

Rails 8 and PostgreSQL, Solid Queue for the runs, no Redis. Put it on our cloud or your own iron. The automations are data, and data exports.

PORTABLE

Definitions are JSON

Export an automation, import it into another workspace or another installation. Templates ship the same way.

SELF-HOSTED

On hardware you control

Run the platform and the models on premise when the data cannot leave. Self-hosted providers are classified INTERNAL by default.

RING-FENCED

Per workspace, per model

Credentials, model connections and context are workspace assets. One workspace on the private model, the next on the frontier, on one platform.

Next step

Bring the process
everybody dreads.

We map it, run it on SynthWise, and hand you the receipts. If it works, the next process is mostly configuration.

SYNTHWISE · NETHERLANDS / UK Contact
AI-NATIVE PRIVACY-FIRST SELF-HOSTABLE SW·001