Automations you can hand to an auditor.
Build the flow on a canvas, run it durably, and require a person for anything with consequences. Every run keeps its steps, its inputs and its outputs.
- Versioned graphs
- Approvals before effects
- Hash chained run trail
Built where you can see it
Steps on a canvas: models, branches, loops, code, data, knowledge and other agents. Every version is kept, and you can compare two of them before publishing.
Runs that survive a restart
A run is a durable record with checkpoints, not a process holding state in memory. It can pause for an approval, wait for a schedule, and continue where it stopped.
Nothing reaches the outside world unsupervised
Tools are registered with the effect they have. A run receives a capability that is good for that one call, and the broker performs it, so a credential never reaches the model or the flow.
Triggers that fit the work
A schedule, a form, a webhook, a chat message, a task, or a change in a collection. The same flow can answer all of them.
What an auditor gets
- An append only trail of every step, hash chained per run
- An evidence pack for a single run, ready to hand over
- Who approved which effect, and when
- The exact graph version that ran
Next to the tools you already know
Every cell about another vendor is a fact that vendor publishes itself, linked to the page it came from. Nothing here claims they lack a feature, because the point is the shape of the two offers, not a scoreboard.
| What you are comparing | NeuriflyApptivity | n8nn8n | MakeMake | ZapierZapier | OpenClawOpenClaw | HermesNous Research |
|---|---|---|---|---|---|---|
| A seat that carries a usage limit worth more than it costs | Confirmed14 euro per seat a month carries a 20 euro monthly usage limit, about seven times what an ordinary person consumes. Past it work carries on and is charged from a balance you top up, with a budget per API key. One person can stay on usage only for ever. | In partPer workflow execution with unlimited users: 20 euro a month for 2,500 executions, 50 for 10,000, 667 for 40,000.source | In partPer credit, where one module action is a credit: 1,000 a month free, then 9 dollars a month for 10,000.source | In partPer task: 100 a month free on one seat, Professional from 19.99 dollars a month billed annually.source | In partMIT licensed and self hosted, so the software costs nothing and you pay your own model provider.source | In partFree and open source under the MIT licence; you pay your own model provider.source |
| Runs in the European Union without changing plan | ConfirmedEuropean Union by default, on every plan. There is no region to switch on. | In partSelf hosting is offered, so the instance runs wherever you put it.source | In partMake publishes its regions and data handling on its own pages.source | UnconfirmedSee Zapier's own pages for where data is processed.source | ConfirmedIt runs on hardware you choose, so the location is yours.source | ConfirmedIt runs on hardware you choose, locally or in Docker, so the location is yours.source |
| Models from more than one vendor on one account | ConfirmedModels from several vendors, including open weight ones, picked per conversation and billed on one account. | In partConnect whichever provider you like and pay that provider separately.source | In partConnect whichever provider you like and pay that provider separately.source | In partConnect whichever provider you like and pay that provider separately.source | In partYou bring your own model keys and pay each provider yourself.source | In partYou bring your own model keys and pay each provider yourself.source |
| Every call recorded with its tokens, latency and cost | ConfirmedAn append only trail with the model, tokens, latency, cost and the key that made the call. Agent runs are hash chained. | In partExecution history is kept in the instance you run.source | In partScenario history is kept in the account.source | In partTask history is kept in the account.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source |
| A budget per key, refused before the spend | ConfirmedA limit per key and per organization. The call is refused before it goes past the limit rather than invoiced after. | In partThe plan caps how many executions a month are included.source | In partThe plan caps how many credits a month are included.source | In partThe plan caps how many tasks a month are included.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source |
| Runs on a schedule | ConfirmedA schedule trigger, run by the platform's own scheduler rather than by a process that has to stay awake. | ConfirmedA standard part of the product.source | ConfirmedA standard part of the product.source | ConfirmedA standard part of the product.source | In partYou schedule it yourself on the machine that hosts it.source | In partYou schedule it yourself on the machine that hosts it.source |
| Webhook and form triggers | ConfirmedWebhooks, forms, a chat message, a task or a change in a collection can all start a run. | ConfirmedA standard part of the product.source | ConfirmedA standard part of the product.source | ConfirmedA standard part of the product.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source |
| A large catalogue of ready connectors | In partTools are imported from an MCP server or an OpenAPI description and approved by a second person, rather than picked from a catalogue of thousands. | ConfirmedA large catalogue of ready integrations.source | ConfirmedA large catalogue of ready apps.source | ConfirmedThe largest catalogue of connected apps in the category.source | In partAround fifty integrations at the time of checking.source | In partDozens of built in tools and a long list of messaging channels.source |
| A person approves before anything irreversible | ConfirmedA step that would do something irreversible waits in an approval centre, and the decision is part of the run's record. | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source |
| Each run is hash chained with an evidence pack | ConfirmedThe steps of a run are chained so a later edit is detectable, and the pack collects what the run read, wrote and spent. | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source | UnconfirmedSee the vendor's own pages.source |
| A code step that runs with no network of its own | ConfirmedCode runs in the sandbox with its network cut by a filter the runner installs on itself, and the run is billed and recorded per second. | In partA code step runs in the instance you host.source | In partCustom code is available on the higher plans.source | In partA code step is part of the product.source | In partIt executes commands directly on the machine that hosts it.source | In partIt executes on the machine that hosts it, through its own terminal backends.source |
Confirmed
In part
Unconfirmed
Last updated 4 August 2026
What your team gets
Five parts, one account, one set of permissions.
Chat and Fusion
Ask across models, grounded in your own documents, with citations you can open.
Knowledge
Your documents, searchable by meaning, with access granted per base and per group.
Code and Data
Functions, a document store and hosted sites, without leaving the platform.
API
An OpenAI compatible endpoint with a budget per key and a receipt per call.
Routing
One name in front of your models, and rules that pick which one answers each call.
Automate one thing that annoys everybody
Start with a weekly report or an inbox that needs triage, and put an approval in front of anything that sends.