Publish the incidents
The status page shows outages, not uptime percentages rounded in our favour. A reliability product that hides its own failures is asking for trust it has not earned.
About
We got tired of agents that quit halfway.
Apkar started as internal infrastructure. It became a product when four other teams asked to use it.
How we got here
We were building an internal agent to triage support tickets. It worked in testing and stalled constantly in production — not because the model was wrong, but because a tool would go quiet and nothing was watching. The only recovery was to start the ticket over by hand.
So we wrote checkpointing, retry policies, and an audit trail around it. That scaffolding turned out to be more valuable than the agent. Four other teams asked to use it before we had a name for it.
We are not building a model. There are excellent ones, and we would like our customers to move between them freely — which is easier when the execution layer belongs to none of them.
How we work
Ship the boring parts
Checkpoint ordering and idempotency keys do not demo well. They are also the difference between an agent you can put on call and one you cannot.
Arabic is not a translation
Our documentation and product are built in both languages by people who write in both. What should not simply be translated gets rewritten.
Say when it does not fit
We tell teams their use case is a poor fit while the deal is still open. It costs revenue occasionally and saves a bad quarter regularly.
The team
Lina Sarwat
Co-founder
Omar Haddad
Co-founder, Infrastructure
Rania Fahmy
Engineering
Nour Abdelaziz
Design Research
Dana Okonkwo
Platform
Yusuf Barakat
Security
Backed by
- Ardent Capital
- Kite Ventures
- Sabil Partners
- North Fund