The black box
You get an answer, but not a usable record of how the system got there. When the result is wrong, it is hard to inspect the decision, find the bad input, or explain the outcome to someone else.
Hard to inspect The agent platform you can actually see into
See every decision your agents make.
Chanly turns complex work into visible agent workflows. Define the outcome, inspect the plan, watch agents and tools execute, and step in when judgment is needed. You should never have to guess what the system did.
Scroll to enter the system
01
Goal
Describe the outcome, constraints and what success should look like.
02
Plan
Review the steps, agents, tools and checks before execution starts.
03
Execute
Follow the run as work moves across agents and connected systems.
04
Learn
Keep the outcome, corrections and useful context for the next run.
01 — The fracture
Most AI work happens somewhere you cannot look.
The problem is not that AI cannot act. It is that once it starts acting across tools and data, teams often lose the thread of what happened and why.
The broken chain
Real workflows cross tools, approvals and people. When one step fails, the failure is rarely isolated: context is missing, ownership is unclear, and the next person has to reconstruct the run by hand.
Hard to recoverThe locked ecosystem
A workflow should not have to be rebuilt because a team changes a model, data source or business tool. The execution layer should stay portable while the components around it evolve.
Hard to change02 — The loop
Goal, plan, execute, learn.
Every run moves through the same four states. The goal is to keep the transition between them explicit enough that a person can understand where the system is, what it is waiting on, and what changed.
Goal
Start with the outcome, not a list of automation steps. Chanly turns the request into a structured objective with context, constraints and success criteria that can be reviewed before work begins.
Plan
The planning layer decomposes the objective into steps, identifies the agents and tools needed, defines dependencies, and surfaces the plan before any side effect occurs.
Execute
Approved steps run with scoped access. The run exposes active state, tool activity, errors and checkpoints so a person can follow or intervene without restarting the workflow from scratch.
Learn
The result, corrections and feedback become reusable context. Future runs can start with what worked before, while changes remain visible instead of disappearing into hidden memory.
03 — Agent network
Agents that work as a system, not a pile of bots.
A useful workflow often needs more than one kind of reasoning. One agent may understand the objective, another may analyze data, another may work inside a business tool, and a person may need to approve the final action. Chanly keeps those handoffs inside one visible run.
The graph below is running a scheduler right now. Colour is the state: blue is the system moving work, violet is a model reasoning, mint is a tool call in flight, amber is a step waiting to be retried.
dispatchreasoning
tool callwaiting
Run graphinterface concept
Agents · live state
Event stream · live
Orchestrator
Owns the objective, decides what work needs to happen, delegates tasks and brings the outputs back together.
Specialists
Focused agents handle bounded jobs such as research, analysis, document review, customer operations or engineering tasks.
Shared run state
Context, outputs, approvals and errors stay attached to the same run so the next step does not have to guess what happened upstream.
05 — The console
The graph is the product.
The console is where a run becomes legible. Follow the plan, open a step, inspect the tool activity behind it, see where an approval is waiting, and understand how the workflow reached its current state.
Interface concept · early product view
06 — Where we are
What works today — and what we are building next.
Chanly is early. Instead of publishing aspirational customer metrics, we would rather show the product surface as it becomes real.
Ready when you are
Open the box.
Bring us one workflow that feels too important to hand to a black box. We will show you how we think about the goal, the plan, the agents, the tools and the human checkpoints around it.