FlowGenX Companion
Build the platform
by describing it.
Companion is the conversational control plane for FlowGenX. Describe the outcome you want and it designs the workflow, connects the systems, runs the test, and ships it to production — pausing for your approval before anything consequential happens.
The gap
Chat that writes config is a demo.
Chat that ships to production is a control plane.
Most build-by-chat tools stall on the same three problems. Companion was designed around them rather than after them.
It can't reach your systems
A model that can only write config still needs a human to wire credentials, resolve endpoints, and handle OAuth. The conversation stops exactly where the real work begins.
It can't be trusted with production
An assistant that can publish a workflow can also publish the wrong one. Without a governance model, the only safe answer is to give it read-only access — which makes it a search box.
It can't show its work
A wall of text is not an audit trail. When something goes wrong, teams need to see which tool ran, with what arguments, who approved it, and what it changed.
What makes it different
Four things Companion does that a chatbot cannot
Governed by construction
Every tool Companion can reach declares a risk tier and a reversibility envelope before it is ever offered to the model. Approval isn't a setting you remember to switch on — it's part of the tool's definition.
- Four tiers, from read-only to destructive
- Reversal procedure shown before you approve
- Justification required for destructive actions
Every system becomes a tool
Companion assembles its toolset per conversation from your managed connectors, your internal services, and any MCP server your team has connected — so it can act across the estate instead of describing what you should do.
- 250+ managed enterprise connectors
- Internal APIs exposed as native tools
- Connect an OAuth app without leaving chat
Conversation to production
The same session that drafts a workflow refines it on a live canvas, saves it, and publishes it. There is no export step, no hand-off to a different builder, and no copy-pasting between tools.
- Generate a workflow graph from a description
- Refine it in place as you talk
- Save and publish through a gated lifecycle
A surface, not a chat log
Companion replies in structured components — tool cards, approval prompts, credential forms, connect buttons, sortable result tables, and a workflow canvas — rendered inline in the conversation.
- Tool activity visible as it happens
- Results as tables you can sort and act on
- Live workflow preview beside the thread
Under the hood
What happens when you hit enter
One turn, five stages. The governance checkpoint sits between the model and your systems — not in a policy document.
You describe an outcome
In plain language, in an authenticated session scoped to your tenant and environment.
The toolset is assembled
Connectors, internal services, connected MCP servers, and workflow tools are gathered for this turn. Anything your admin disabled is removed before the model sees it.
The model plans and acts
Your organisation's model of choice reasons over the available tools and starts calling them, streaming its thinking as it goes.
Risk tiers gate execution
Reads and safe writes run immediately. Consequential and destructive actions stop and wait for a human decision.
You see every step
Each tool call, result, approval, and form arrives as its own card in the thread — not buried in prose.
Human in the loop
Autonomy you grant, one tier at a time
Risk classification is a property of the tool, not a prompt instruction the model can talk itself out of. The gate sits in the execution path.
Listing connectors, searching for tools, reading a record, inspecting a workflow. No side effects, no friction.
Drafting a workflow graph, saving a draft, creating a scratch record. Reversible by design and captured in the audit trail.
Publishing a workflow, writing to a production system, sending an external message. Execution pauses until a human approves it in the thread.
Deleting data, revoking access, tearing down a deployment. The approver must record why before the action is allowed to proceed.
Approvers see the consequences, not just the command
Every gated action declares whether it can be undone.
The reversal procedure is shown to the approver up front.
The estimated impact of leaving it in place is stated plainly.
Tool fabric
Companion is only as useful as what it can reach
Its toolset is assembled per conversation from four sources, then filtered by your policy — a single, auditable path to every system it can touch.
Managed connectors
Over 250 enterprise systems — CRM, ITSM, HRIS, finance, storage, messaging — available as governed tools with credentials resolved per tenant.
Your internal services
Bring an existing API and it becomes a first-class tool set. No bespoke connector development to make internal systems reachable.
Connected MCP servers
Teams can attach their own MCP servers; their tools join the same assembly path and inherit the same tiering and audit behaviour.
OAuth without leaving chat
When a system needs authorising, Companion renders a connect card in the thread, waits for the handshake, and resumes the task.
Administrators decide the surface
Tools disabled for a user are subtracted before the toolset reaches the model — so a disabled capability is genuinely unavailable, not merely discouraged.
Results you can act on
Records come back as sortable tables with row actions, so the next instruction is a click rather than a re-typed query.
Workflow lifecycle
From a sentence to a published workflow
One session carries the whole lifecycle. Nothing is exported, re-imported, or rebuilt in a second tool.
Describe
State the outcome in a sentence. Companion asks only for what it genuinely cannot infer.
Generate
A workflow graph is drafted and streamed onto a canvas beside the conversation as it forms.
Refine
Change it by talking — add a branch, swap a connector, adjust the schedule. The canvas updates in place.
Save
The draft is persisted against your tenant. Reversible, logged, and available to open in the full builder.
Publish
A tier-gated action. Companion pauses, shows what publishing will affect, and waits for your approval.
Prefer the canvas? Everything Companion builds opens in the full agentic workflow builder — same graph, same runtime, no translation layer.
Who it's for
One surface, four different jobs
Automation engineers
Ship the first version in the time it used to take to scope it.
- Draft a multi-step workflow from a description
- Find the right connector endpoint without reading docs
- Test an execution and diagnose the failure in the same thread
Platform & operations
Operate the estate conversationally, with a record of every change.
- Inspect deployments and configuration on demand
- Make production changes behind an approval gate
- Keep an auditable trail of who approved what
Security & governance
Decide what AI is allowed to do — and prove it afterwards.
- Set which tools each user's assistant can even see
- Require justification before destructive actions
- Review the reversal path prior to approving
Business teams
Describe the process you run; get the automation that runs it.
- Build without learning the canvas first
- Connect the apps you already use
- Hand the result to IT as a governed workflow
Security & governance
Built for the team that has to sign it off
The governance controls are structural. They sit in the path between the model and your systems, which means they hold regardless of what the model decides to attempt.
Authenticated, tenant-scoped sessions
Every Companion session is bound to an authenticated user, tenant, and environment. Identity is established at connection time, not asserted by the client.
Credentials the model never sees
Connector secrets are resolved inside the execution layer at call time. Tools receive a reference; the model receives a result.
Per-user tool visibility
Administrators control which tools reach which users. Filtering happens where the toolset is assembled, so it cannot be prompted around.
Model choice stays yours
Companion routes through your organisation's model configuration, so provider and model selection remain a platform decision.
FAQ
Common questions
What is FlowGenX Companion?
FlowGenX Companion is the conversational control plane for the FlowGenX platform. You describe an outcome in natural language and Companion designs the workflow, connects the required systems, runs tests, and publishes to production — calling real platform tools rather than only generating configuration for you to apply yourself.
How does Companion prevent an AI assistant from making unsafe changes?
Every tool Companion can call declares a risk tier as part of its definition. T0 read operations and T1 safe writes run immediately and are logged. T2 consequential actions, such as publishing a workflow or writing to a production system, pause and wait for explicit human approval. T3 destructive actions require both approval and a written justification. Because the tier lives on the tool rather than in the prompt, the model cannot negotiate its way past the gate.
What does an approval request actually show me?
Before you approve a gated action, Companion shows the tool being called, whether the action is reversible, the procedure for reversing it, and the estimated impact of leaving it in place. Approvers judge the consequence, not just the command name.
Which systems can Companion connect to?
Companion assembles its tools from over 250 managed enterprise connectors, your own internal APIs exposed as tools, and any MCP servers your team has connected. When a system needs authorising, Companion presents an OAuth connect card inside the conversation and resumes the task once the handshake completes.
Can administrators limit what Companion is able to do?
Yes. Administrators control tool visibility per user, and disabled tools are removed at the point where the toolset is assembled — before it reaches the model. A disabled capability is genuinely absent from the session rather than merely discouraged by instructions.
Does Companion replace the visual workflow builder?
No. They are two views of the same artifact. Companion streams the workflow graph onto a canvas as it builds, and anything it produces opens in the full agentic workflow builder with the same graph and the same runtime. Teams move between conversation and canvas without a translation step.
Which AI models does Companion use?
Companion routes model calls through your organisation's configured model layer, so provider and model selection stay a platform decision rather than something hard-coded into the assistant. FlowGenX is model-agnostic and supports commercial and open-source models.
Describe a workflow.
Watch it reach production.
Bring a real process from your business. We'll build it live in Companion — connectors, approvals, and publish included.