Build AI customer support agents you can actually trust.
FlowAssist is not another helpdesk. It is the toolkit for building the AI agents that answer inside the helpdesk you already run: Chatwoot today, Freshdesk and Zendesk next, or your own in-house system through a custom integration. It brings software-engineering discipline to those agents: versioned policies, tests before you ship, one-click rollback, and a full audit trail. Some of it works today, some is still on the drawing board. Have a look, then tell us what your team needs.
We reply personally, usually within two business days. No newsletters unless you ask.
Maya R. · It has been 6 business days and I still do not see the refund…
Negative
Refund96%Payment delay81%
How do I add a teammate?
Jonas K. · Loving the product so far. Is there a way to invite my colleague…
Positive
How-to93%Seats77%
Export to CSV is empty
Priya S. · The export downloads but the file only has a header row…
Neutral
Bug89%Export84%
Refund policy
v3 → v4
Refund policy
- Refunds are approved for orders under 14 days old.
+ Refunds are approved for orders under 30 days old.
+ Escalate to a human when the customer mentions a chargeback.
Always confirm the order number before acting.
Illustrative preview. Names and numbers are made up.
Built for
Support teams with 5–50 agents
Support Ops leads
Agent builders who do not write code
Enterprise support & IT teams
Most AI support agents are built like it is 1999.
Teams are handing customer conversations to models with none of the safeguards they would demand from any other production system. That is the problem we are working on.
No versioning
A prompt gets edited in a text box and the old behaviour is gone. Nobody can say what the agent believed last Tuesday.
No tests, no rollback
Changes go straight to real customers. When quality drops, the only fix is to guess and edit again.
No audit trail
Who changed the refund policy? Which model answered that ticket, with which keys? Most tools cannot tell you.
What works today
Everything you need to build agents that work inside your helpdesk.
FlowAssist is not a support platform. It is the layer that builds, versions and ships AI agents into the tools you already run. These four pieces work on real conversations right now. Everything marked “roadmap” further down does not, yet.
Understand every conversation
Keep your helpdesk. FlowAssist plugs into it, pulls in real conversations and classifies every ticket automatically.
Tag classification with confidence scores
Suggestions for tags you have not created yet
Sentiment analysis on every incoming message
Chatwoot today; Freshdesk and Zendesk next; custom helpdesks via signed webhooks and our API
Policies you can version and roll back
Write support policies in plain language. Every save is a version you can diff, compare and restore.
Plain-language policies, no prompt engineering degree required
Side-by-side diffs between any two versions
Roll back to a known-good version in one click
Replay and regression gates on the roadmap
Connect your APIs and MCP servers
Give the agent the tools your human agents already have. Order lookups, refunds, account changes, all behind your rules.
Register your own REST APIs as tools the agent can call
Connect any Model Context Protocol (MCP) server and pick which tools to expose
Every tool call is logged with inputs, outputs and the policy version in force
Per-tool permissions so sensitive actions can require human approval
Your models, your keys, your data
Use Anthropic Claude, OpenAI GPT or Google Gemini with your own API keys. No model lock-in.
Switch providers without rewriting policies
Keys encrypted with cloud KMS (AWS, GCP, Azure, Vault)
Full audit log of every configuration change
Multi-tenant isolation designed in from day one
How it will work
Five steps from the helpdesk you already have to an agent you can safely ship into it. Three work today, two are still being built.
Step 1
Connect
Plug into the helpdesk you already use, or a custom one, plus your APIs and MCP servers.
Step 2
Understand
Tickets are tagged, scored and sentiment-checked automatically.
Step 3
Author
Write policies in plain language. Every save becomes a version.
Roadmap
Step 4
Test
Replay real conversations against a candidate config before it goes live.
Roadmap
Step 5
Ship
Promote with confidence, watch quality, roll back in one click.
Roadmap
What works today, what is next, and where we want to go. Design partners move items between these columns.
Now
What already works
Chatwoot integration ingesting real conversations
Custom helpdesk integration via signed webhooks and API
Automatic tag classification with confidence scores and new-tag suggestions
Sentiment analysis on incoming messages
Versioned, diffable, rollback-able support policies
Self-hosted deployment in your own cloud for enterprise teams
Next
What we intend to build next
Playground to chat with a candidate configuration
Replay real conversations against a new policy version
Freshdesk and Zendesk connectors
Guided setup for custom and in-house helpdesk integrations
Later
Where we want to go
Regression eval gates that block risky releases
One-click rollback driven by live quality signals
Team review flows for policy changes
Deeper analytics across tags, sentiment and outcomes
Design partners
Build it with us, on your real tickets.
We work with a handful of support teams who use the unfinished product and tell us where it falls short. If that sounds like your team, introduce yourself.
Shape what gets built
Design partners decide which connectors, models and workflows we build next. Your backlog becomes our backlog.
Hands-on onboarding
We connect your helpdesk with you, tune the first policies together and stay in the loop while it runs.
Use it while it is unfinished
Run the parts that work today on real conversations, with the founders watching quality alongside you.
A direct line to the founders
A shared channel with the people building the product, not a ticket queue.
Enterprise
Run it inside your own walls.
Self-host FlowAssist in your cloud, encrypt with your KMS, connect your internal APIs and MCP servers, and keep a complete audit trail of everything the AI does.
Self-hosted in your cloud
Run FlowAssist inside your own AWS, GCP or Azure account, or on-prem. Customer conversations never leave your perimeter.
Your KMS, your keys
Secrets and model keys are encrypted with the key management service you already trust: AWS KMS, GCP KMS, Azure Key Vault or HashiCorp Vault.
Complete audit trail
Every policy edit, model change, tool call and rollback is recorded with who, when and which version. Export it to your SIEM.
Tenant and team isolation
Brands, regions or business units get isolated workspaces with their own policies, keys and connectors.
Integrates with your stack
We are not a helpdesk. Connect Chatwoot today, Freshdesk and Zendesk next, or your in-house system through a custom integration, then add internal APIs and MCP servers as tools.
Data residency by design
Self-hosting means you decide the region and the retention. No shared infrastructure, no cross-tenant data.
Talk to us about a self-hosted deployment
Tell us your cloud, helpdesk and the systems the agent should be able to reach. We will come back with a deployment plan.
Frequently asked questions
Is FlowAssist finished?
No, and we say so on purpose. Ingestion, classification, sentiment, versioned policies and bring-your-own-model work today. Testing, replay and one-click rollback are still being built. This page shows both so you can judge for yourself.
Can my team use it now?
A small number of design-partner teams run the working parts on real conversations with us. Tell us about your setup and we will let you know whether it is a fit.
Is FlowAssist a helpdesk or a customer support platform?
No. FlowAssist helps you build, version, test and ship the AI agents that work inside your existing support tools. Your team keeps its helpdesk, inbox and workflows; the agent shows up there.
Which helpdesks does it integrate with?
Chatwoot works today. Freshdesk and Zendesk are next. If you run something else, including an in-house system, we integrate with it through signed webhooks and our API, and we will do that setup with you.
Which AI models can I use?
Anthropic Claude, OpenAI GPT and Google Gemini, using your own API keys. You can switch providers without rewriting your policies.
Can we self-host FlowAssist?
Yes. Enterprise teams can run FlowAssist in their own AWS, GCP or Azure account or on-prem, using their own KMS for secrets. Tell us about your environment and we will walk you through the deployment.
Can the agent talk to our internal systems?
Yes. You can register your own REST APIs and connect MCP servers as tools. You choose which tools the agent may call, and every call is logged against the policy version that authorised it.
How are my data and API keys protected?
Keys and secrets are encrypted with a cloud KMS (AWS, GCP, Azure or Vault). Every configuration change is written to an audit log, and tenants are isolated from each other by design.
Is there a free trial?
Pricing and trial details will be announced at launch. Design partners agree terms with us directly before then.
How do I get in touch?
Leave your work email in any form on this page and a founder will reply. If you prefer, mention your helpdesk and team size so we can prepare.
See it for yourself, unfinished edges included.
Leave your work email and a founder will get back to you with a walkthrough of what works today and what is coming next.