Support agent
Answers enquiries, status checks and policy lookups, and escalates to a human when needed.
Conversational automation, without the spaghetti
Turn business processes into conversational agents.
Describe the customer journey you want to automate. AvlyFlow helps you build, test and deploy AI-powered agents on WhatsApp, Web, SMS, USSD and Voice — with payments, APIs, knowledge and human handoff built in.
Self-hostable · API-first · PostgreSQL + pgvector · S3-compatible storage
4channels, one flow
Minutesto a working agent
24/7automation with handoff
APIeverything the UI does
ENTRYwelcomeSend welcome message
→
STANDARDcollect_detailsValidate & save input
→
AI ACTIONqualify_leadLLM + tools
→
TERMINALcompleteDeploy anywhere
The problem
Customers already message you. The work is still manual.
Answering the same questions, collecting the same details, sending the same payment links, chasing the same orders. It works at 20 conversations a day, breaks at 200, and is impossible at 2,000.
The solution
Automate the routine. Keep the human for what matters.
AvlyFlow turns those repeated journeys into agents that run 24/7 across every channel your customers already use — while your team handles the complex cases through built-in human handoff.
Use cases
Six agents every business should have
Start from a proven outcome, not a blank canvas.
Booking agent
Collects details, checks availability, confirms appointments and sends reminders.
Payment collection
Triggers payments inside the conversation, confirms receipt, and sends the confirmation.
Survey agent
Asks structured questions, records responses, and follows up — in Swahili, English or mixed.
Ticketing agent
Sells tickets end-to-end: select, pay, and receive the ticket in one conversation.
Lead qualification
Collects budget, location and need, then routes the qualified lead to your sales team.
How it works
Describe it. Build it. Test it. Deploy it.
From idea to a live agent in four steps.
1
Describe it
Tell the AI copilot what process to automate. Support, bookings, payments, surveys — start from the outcome.
2
Build it
Design the journey visually. Add AI responses, connect your knowledge base, wire payments and API calls. No code for core journeys.
3
Test it
Preview the agent as a real conversation across every channel before going live. Catch gaps and edge cases.
4
Deploy it
Publish to WhatsApp, Web, SMS, USSD or Voice. Connect APIs, payments and handoff, then monitor in real time.
One runtime, many patterns
Deterministic when it matters. Agentic when it helps.
Use explicit states for payments, forms and compliance-sensitive steps. Drop in an LLM agent for discovery, support or complex tool use. Mix both in the same workflow.
Visual state canvas
Drag states, connect transitions, edit JSON, preview execution and inspect traces.
AI workflow copilot
Describe what the chatbot should do. Generate a valid flow, edit it conversationally, then test before publishing.
Knowledge with pgvector
Store embeddings directly in PostgreSQL. Attach sources and retrieve scoped chunks for agents.
Payments as adapters
Use one payment interface for ClickPesa, AzamPesa, DPO and your own providers.
Channels & integrations
Web chat, WhatsApp, SMS/USSD and Voice share one execution model, with REST APIs, MCP tools and human handoff underneath.
Built for operations
Tracing, metrics, audit logs, idempotency, retries, rate limiting and versioned publishing are first-class.
Channels
One flow. Every channel your customers already use.
Design the journey once and deploy it everywhere — no channel duplication.
WebEmbeddable widget
WhatsAppBusiness API
SMSTwo-way
USSD*123# menus
VoiceIVR + speech
REST / MCPProgrammatic
For developers
API-first under the hood. Your stack stays yours.
The visual builder is one client of the same REST API you get. Script agent creation, keep flows in git, wire conversations into your own services.
- REST API for chatbots, flows, conversations, knowledge and analytics
- Webhooks and
API_CALLnodes to connect any backend mid-conversation - MCP servers per chatbot for tool-using AI agents
- Flows are plain JSON — edit them in the built-in IDE, keep them in git
curl -X POST http://localhost:8787/v1/chatbots/{id}/chat/sessions/{sid}/messages \
-H "Authorization: Bearer sk_avlyflow_..." \
-d '{"message":"What are your hours?"}'
# → { "session_id": "...", "messages": [ { "type": "text", "text": "..." } ] }
Deploy locally in minutes
Your data plane can stay yours.
The default stack uses PostgreSQL/pgvector, Redis and MinIO. Moving object storage to AWS S3 only requires configuration changes because MinIO speaks the same S3 API.
docker compose up --build # app http://localhost:8788 # minio http://localhost:19101 # grafana http://localhost:3334
AvlyFlow App
↓
PostgreSQL + pgvector
Redis
MinIO / S3
Prometheus + Grafana
Common questions
Answers, before you ask.
Is AvlyFlow just another chatbot builder?
No. A chatbot mainly replies to messages. AvlyFlow agents can answer, collect data, trigger payments, call APIs, run workflows and escalate to a human — a conversational business-automation platform that runs across WhatsApp, Web, SMS, USSD and Voice.
Can it integrate with our existing systems?
Yes. The API_CALL action makes outbound requests to your backend, and webhooks push events back to you. Customer records stay in your system of record.
Can our engineers build on it programmatically?
Yes. Everything the UI does is available over the REST API, and flows are plain JSON. Use the OpenAPI reference and MCP server.
Is it self-hostable?
Yes. Run the full stack locally with Docker Compose, or deploy to your own infrastructure.
Build the conversation. Own the runtime.
Start with the complete local stack, then swap infrastructure providers without changing your workflow DSL.
Start with AvlyFlow