01
Graph-first AI agent builder
Build AI agents as workflow graphs — typed nodes, conditional edges, versioned drafts, one-click publish.
AI agent platformBuild AI agents visually.
Build AI agents visually.
The agent platform for production agentic AI.
Build agents you can see. Design each AI agent as a workflow graph, connect tools, knowledge and memory, run multiple agents behind one app, and watch every run execute node by node with nothing hidden.
Graph-first builderSemantic tool retrievalpgvector RAGScoped memoryHuman hand-offSlack and webhooksFull run traces
Agent 61 · Run 1352Graph v3
Inputsession 497
Memory read94 ms
Knowledge6 chunks
LLM3,115 tok
Toolbox11 tools
Finalizestream
Node executionsIdle
The run, end to end
Design agent workflows visually
Scroll: the graph executes. This is the actual shape of an AgentRow run, the same graph you build in the editor, in the same order the runtime walks it.
Live graph · scroll to executeStage 01/06
Agent graph · v3Executing
01/06
Input
A message arrives from chat, API, or an app event. Session, attachments, and app context enter the graph.
One workspace
Build, test, and trace in one place
Chat with the agent while you edit it. The workflow, tools, knowledge, memory, and logs live beside the conversation.
AgentRow · Visual agent builder
AgentRow · App workflow · Channels
AgentRow · App workflow · Support hand-off
Apps
Orchestrate multi-agent systems
An app puts multiple agents behind a single surface. Its own graph routes each message to the right agent, runs it, screens the answer, and every nested run is traced like any other.
01Stage
Router
Selects the agent for each message from the attached agents' descriptions and roles.
02Stage
Nested agent run
Runs the selected agent's published workflow with the conversation and the user's context.
03Stage
Chat (out) with safety rules
Screens the selected agent's answer with the app's own output rules and streams it to the chat surface.
04Stage
Widget, API, Slack, webhooks, or your team
Reaches users through the chat widget, the end-user API, a Slack bot or a signed webhook; a support app can hand the conversation to a person in the inbox.
Support
Hand off to a person, on the record
A support app gives its agent one more tool: hand this conversation to your team. The customer keeps the same chat, your team replies from the inbox, and the hand-off sits in the run trace like any other step.
01Support
The agent decides to hand off
Attach the Support hand-off node to an Agent node. You write when it should hand off and how it should reply, in plain language; the agent gets a built-in tool that does exactly that.
02Support
Your team replies from the inbox
Open and Resolved tabs per app, the agent's reason and summary on every hand-off, and replies that reach the customer's chat under the team member's name.
03Support
Nothing hidden
The hand-off is a tool call in the run trace. A resolved conversation reopens when the customer writes again, and the AI stays out of the thread until then.
Now and next
Where the platform is, and where it is going
What you can use today, what is being built for the coming months, and what comes after. Dates move; the order does not.
Available now
- Visual agent builder, tools, knowledge bases, memory
- Apps on the chat widget, the end-user API and Slack
- Signed inbound webhooks, server runs and webhook delivery
- Support apps: hand-off node, inbox, replies by name
- Full run traces, usage and cost analytics, access tokens
- Bring your own model keys
Next few months
- WhatsApp and Telegram as channels of the same app
- Inbox replies delivered back into Slack
- Shopify app: install the assistant from the App Store
- Marketplace: paid listings, ratings and public listing pages
- Support roles and seats, hand-off notifications
- Plans and billing, with Row, the model we provide
After that
- Sell agents and apps: subscriptions and one-time purchases
- Gmail as a channel
- AI-drafted replies for the team to edit
- WooCommerce plugin
- Agency and multi-store management
The marketplace
Share and install AI agents and apps
Share a tool, an agent or a whole app with other organisations, free, and install what others shared as your own copy in one step. Selling comes later.
01In the product today
Share and install
Share a tool, a toolbox, an agent or a whole app, free. A public listing installs as an editable copy, a private one as a sealed copy others can use but not open. Secrets never travel: the installer fills them in.
02In the product today
Updates in place
Every version is reviewed before it goes live. When you publish a new one, installers are told and update when they choose: their answers are kept, what they changed is theirs to keep or take, and a changed workflow arrives as a draft.
03Planned
Sell installs
Planned: charge a subscription or a one-time price. Buyers install a sealed copy of the version you published, configure it and run it without ever seeing your workflow.
Channels
One app, every channel
The same published graph answers on your website, in Slack, from a signed webhook or a server call, and can post every answer to your own endpoint. Channels are nodes on the canvas: where a turn came from and where the answer went are part of the run.
01Channel
Chat widget API
An embeddable chat widget with a JavaScript API (init, identify, open, events) and a headless client, or your own UI on the end-user API. Turns enter through Chat (in), the one place for context and input rules.
02Channel
Slack bot
Your own Slack app, connected in a minute. Anyone in the workspace DMs or mentions it; answers land in the thread, and every thread is in the Inbox.
03Channel
Webhook (in)
Other systems start runs by POSTing signed JSON. Signatures checked before parsing, duplicates recorded and never re-run, every event listed.
04Channel
Server runs
Your backend starts a run with the app secret: thread keys, sync or async with a callback, idempotency, optional identity.
05Channel
Webhook (out)
Every finished answer posted to your endpoint as signed JSON. A delivery is a traced step; a failed one never fails the run.
Platform
Everything you need to build AI agents for production
02
Semantic tool retrieval
The toolbox picks the right tools per message with vector search plus lexical reranking — no prompt stuffing.
03
Tool and API integrations
Give agents access to APIs, databases, webhooks, internal services, and custom actions.
04
Knowledge and RAG
Ground responses with document knowledge bases, pgvector retrieval, and score-aware reranking.
05
AI memory, scoped
Workspace, app, agent, and user memory scopes with explicit read and write nodes in the graph.
06
Runtime observability
Inspect traces, tool calls, logs, tokens, latency, errors, and step-by-step execution.
07
Support inbox and hand-off
The agent hands a conversation to a person; your team replies from the inbox and the customer sees it in the same chat.
08
Production controls
Add approvals, guardrails, secrets, rate limits, and deployment-ready execution rules.
Observability
Observe every run, on the record
A clear execution timeline for every run: model calls, tool inputs and outputs, retrieval scores, trims, errors, latency, and token cost. No more guessing why an agent behaved a certain way.
Node-by-node waterfall with timings
Tool selection scores and drop reasons
Context trims and token budgets
Errors, retries, and cost per run
Run 1352 · Node executionsCompleted
Input<1 ms
Memory read94 ms
Knowledge retrievalTrimmed135 ms
LLM809 ms3,115 tok
Toolbox233 ms
LLM3.14 s4,204 tok
Router<1 ms
Context builder<1 ms
Finalize<1 ms
Memory writeskipped
Duration
5.85 s
Tokens
8,327
Steps
11
Tool calls
1
AI agent examples
Agent apps teams run today
01Agent app
AI customer support
Answer from your knowledge base, look up orders with tools, and hand the conversation to a person in the inbox when the agent should.
02Agent app
Internal operations assistant
Wire internal APIs into the toolbox and let the agent handle lookups, updates, and reports, with an audit trail. Ask it in Slack.
03Agent app
Sales proposal assistant
Ground drafts in your catalog and past proposals, keep tone in instructions, and review every generation.
04Agent app
E-commerce product assistant
Search products semantically, check stock and orders through tools, and remember the shopper across sessions. The Shopify app puts the assistant in the storefront chat; its App Store listing is on the way.
05Agent app
Developer debugging assistant
Feed it logs and docs, give it read-only tools, and trace exactly which context produced each answer.
06Agent app
Workflow automation agent
Start it from a signed webhook or a server call, chain tools with routes and approval gates, and post the result back to your system.
Integrations
Your models, your tools, your infrastructure
Bring your own model keys or point AgentRow at your own model server. One app answers on your site, in Slack and on a Shopify store, its tools reach your systems, and retrieval runs on Postgres with pgvector.
ModelsYour own provider keys, or your own model server at its address.OpenRouterOpenAIAnthropicvLLMOllamaChannelsOne published app answers wherever the conversation is.Chat widgetSlackShopify storefrontWebhooks in and outServer runsToolsTyped actions the agent can call, every call in the run trace.HTTP APIsWebhooksBrowserRAG retrieverShopify storeKnowledgeRetrieval on Postgres with pgvector, from the documents you add.FilesWeb pagesAgentRow EmbeddingSelf-hosted embeddings
AI agents, explained
What is an AI agent?
An AI agent, or intelligent agent in artificial intelligence, is software that takes an input, reasons with a language model, uses tools and memory, and acts to reach a goal, often over several steps. In AgentRow an agent is an executable workflow graph: input, memory, knowledge retrieval, the model, a toolbox loop and a final answer, with every step recorded.
What is agentic AI?
Agentic AI describes systems where agents plan and act rather than answer a single prompt: they decide which tools to call, read the results, and continue until the task is done or a limit is reached. AgentRow runs that loop inside the graph, with a tool budget, approval gates for risky actions and a trace of every iteration.
What types of AI agents are there, with examples?
Textbook types are reflex agents that react to rules, model-based agents that keep state, goal-based and utility-based agents that plan toward an outcome, and learning agents that improve from feedback. In products the common examples are customer support agents, e-commerce assistants, internal knowledge assistants, workflow automation agents and coordinator agents that delegate to specialists.
What is a multi-agent system?
A multi-agent system is several agents working on one problem, usually with a router or coordinator that decides which agent handles a request and passes it the context. In AgentRow an app is that system: a router in front of multiple agents, one answer out, every nested run traced.
What is AI memory?
AI memory is what an agent keeps between conversations: facts and preferences about a user, an app or the agent itself. AgentRow stores memory as explicit items with a scope, reads them through a node in the graph, and asks for approval before anything is saved.
Is there a chat widget API?
Yes. The embeddable chat widget is one script with a JavaScript API: init with your app id, identify to sign a user in with a token your backend issues, open, logout, and events for every message, tool call, hand-off and run. The same script exposes a headless client for a custom chat UI, and the end-user API serves your own frontend or backend directly.
Can I build AI agents without code?
Yes. The visual builder covers instructions, model, tools, knowledge bases and memory, and every new agent starts with a working workflow. Developers get the same graph with per-node settings, an end-user API, webhooks and server runs when they need them.
Free trial
Ready to build your first AI agent?
Start with a free workspace, wire your first graph, and scale when your agents do. Product teams and AI agencies building assistants for clients get the same platform, with a Scale plan for many stores. Sign up and the first month is free.