Skip to content
The person behind it

Taha Abdelrahman

Founder & sole engineer, Wasl AI

I build tools that take repetitive thinking off people's plates. Wasl AI is the one I needed most, so I built it first.

28

executable node types

9

models available

3

model providers

52

end-to-end checks

Why Wasl AI exists

Automation tools are very good at moving data and hopeless at forming a judgement. Language models form judgements beautifully and cannot run at six in the morning, reach into your systems, or remember what happened yesterday. Wasl AI lives in the gap between those two facts.

I kept watching capable people spend their best hours on work a machine should have finished overnight — opening the same pages, making the same calls, writing the same summary in slightly different words. That is not a people problem. It is a missing tool.

The bet behind Wasl AI is that the model was never the hard part. The hard part is everything around it: ordering the steps, branching on the answer, looping over a list, recovering from the call that failed, and above all being able to see exactly what happened when the result is wrong.

What I refuse to compromise on

Nothing is a black box

Every node records its inputs, its output, its logs, how long it took and what it cost. If an automation surprises you, you should be able to see why in seconds — not guess.

Your key, your bill

Attach your own model key and those calls cost zero credits. I would rather be paid for the orchestration I actually built than take a margin on somebody else's tokens.

Failure should be legible

A branch that was not taken is marked skipped, not failed. An unreachable page names the host it could not reach. Error messages are written for the person reading them at midnight.

Verified, not assumed

Every release runs an end-to-end suite that boots the built app and exercises it for real. It has caught bugs I would otherwise have shipped, including several found only against the live deployment.

What is actually under the hood

No wrappers, no hand-waving. This is the real inventory.

Execution engineTopologically ordered DAG runner with cycle rejection, automatic per-item fan-out, branch skipping and error routing
Node catalogue28 executable node types across triggers, AI, data, logic, actions and output
Models9 current models across 3 providers, with credit cost derived from list price rather than invented
SecurityAES-256-GCM credential vault, SSRF blocklist on user-supplied URLs, OAuth with PKCE, sandboxed code node
InterfaceEnglish and Arabic with correct RTL, light and dark themes, live run traces streamed over SSE
Confidence52 end-to-end checks covering auth, branching, fan-out, webhooks and the public REST API
3 models3 models3 models

Say hello

I read everything. If you are automating something painful, or you found a rough edge in Wasl AI, I want to hear about it.

See it work

The fastest way to understand Wasl AI is to run something. Pick a template and press Test run.

Open the templates
Connect anything. Automate everything.About · Wasl AI