Your agents work together
You control them in one place
See what your agents do, what they did, and interact with them instantly.
Framework-Agnostic. LangGraph, CrewAI, or any framework. Wraps the agent you already built, wherever it runs. Python SDK, or any language over the REST API.
How it works
See everything your agents do.
AgentWay doesn't ask you to rebuild your agent, adopt a framework, or inherit from a base class. It attaches to the loop you already wrote.
Create a project and a scope
A scope is a department — data,
engineering, whatever fits your team. It gets one
key, and every agent you deploy there uses it.
Change your loop condition
while True becomes
while agent.running(). That single call is your
heartbeat, your pause check, and your inbox — one request per
iteration.
Watch it appear
The agent registers itself on first run and shows up on your canvas, live. No dashboard setup per agent, no configuration file.
from agentway import Agent
agent = Agent(
project_id="aw_x7k2m9p4qa", # from your dashboard
scope="data",
slug="invoice-processor",
name="Invoice Processor",
)
while agent.running(): # heartbeat + pause + directives
invoice = fetch_next()
agent.working(f"processing {invoice.id}")
process(invoice)
agent.done()Direct
Tell an agent what to do. Without redeploying.
Type an instruction and send it to one agent, a whole department, or everything you run. It lands on the agent's next check-in — within seconds.
- Target one agent, a scope, a capability, or the whole project
- See delivered, acknowledged, and acted on as separate states — because they are
- An agent that can't comply says so, with a reason
14:01:30 · you
Sent to data (4 agents): "skip anything from vendor ACME"
14:01:34 · invoice-processor
Acknowledged — "understood, applying now"
14:01:51 · invoice-processor
Acted on — skipped 3 ACME invoices
14:02:02 · ocr-worker
Declined — "no access to vendor metadata"
Approve large invoice?
Which quarter should I report on?
Escalate
Let your agents ask instead of guessing.
An agent that hits something it shouldn't decide alone can stop and ask you. It appears in your inbox, flagged, with the choices it will accept. Answer, and it picks up where it left off.
- Blocking questions sort to the top — you can see who's waiting
- Offer fixed choices, so the answer is safe to act on
- The agent keeps its heartbeat while it waits, so you know it's alive
Oversee
Your whole fleet, arranged the way you think about it.
Agents sit inside the scopes you define, on a canvas you arrange yourself. Status, current task, and trouble are legible from across the room — and every agent keeps a complete, readable history.
- Drag agents where they make sense; the layout is yours and it persists
- Offline detection built in — nobody has to notice a dead process
- Every operator action is written to an immutable audit trail
Syncing entries
Paused by you
Reading ticket #8821
Offline 4m
Our Promise
AgentWay is useful only if you can trust what it tells you.
Here are some facts you may want to know about :
Pause is one click
Your agent stops at its next checkpoint — the top of its loop. For most agents that's a second or two. The dashboard shows "pausing…" until the agent confirms it has actually stopped, so you're never guessing.
Nothing stops mid-action
If your agent is 40 seconds into an API call, pause lands when that call returns. No tool can safely freeze a process mid-write. We show you the real latency instead of pretending it's zero.
Delivery is guaranteed
Every directive is durably stored, ordered, and delivered at least once. You can always see whether an agent received it and what it said back.
Obedience is not
Whether your agent acts correctly on an instruction depends on how you built it. We hand it the message and record what happened; the behaviour is yours.
Our outage isn't your outage
If AgentWay is unreachable, the SDK returns
True and your agent keeps working. A control plane
that takes your fleet down with it would be worse than none.
We don't read your work
Everything is encrypted in transit and at rest, but what your agents report is stored readable — it has to be, to appear on your dashboard. We access it only at your request, to investigate an incident, or where the law requires it.
Stop guessing what your agents are doing.
Create a project, add a scope, drop a key in your environment. Your first agent shows up on the canvas within a minute.