Job description

ABOUT OPENROUTER OpenRouter is the leading AI routing and infrastructure layer that enterprises use to access, manage, and optimize the best large language models across providers—without lock-in, capacity constraints, or unnecessary cost. We power the most advanced AI teams in the world by giving them the flexibility to move fast, scale confidently, and stay future-proof as models evolve. As enterprise adoption of AI accelerates, OpenRouter sits at the center of how organizations operationalize LLMs across research, product, and production workloads. ABOUT THE ROLE We're hiring a Software Engineer on our Ori, to build out our family of products for putting any model to work in your terminal, in Slack, and on your desktop. Ori is built to answer the question from large buyer ICP, we are building the client side products for massive enterprise adoption of OpenRouter beyond the engineering organization. Ori is small team built the current product in a few months. You'll work with that team, which is led by one of our co-founders, alongside product, design and the platform teams Ori is built on. This is a hands-on IC role with broad surface area and little process between you and shipping. WHAT YOU'LL DO - Build and ship across every Ori surface: the ori CLI and its harness integrations. - Make model fungibility real in any harness. That means model switching, provider fallbacks, tool and prompt settings tuned to each model, and respecting org guardrails, all without extra auth for the user. - Build the agent runtime underneath. That includes sandboxed agents that never see their own credentials, secrets management, session management and compaction, and skills that agents can learn, share and inherit across models. - Take agents from a laptop to the cloud: deploy, fork and shut down agents on command, with agent state that is version-controlled and auditable. - Own quality end to end, from install, auth and CLI UX to reliability, observability and evals. Ori is heavily dogfooded internally, so you'll use what you build every day. - Feed the gaps Ori exposes back into OpenRouter's API and SDKs, so the whole ecosystem can build multi-model agents on the same primitives. WHAT WE'RE LOOKING FOR - 5+ years shipping production software, with real ownership of developer tools, CLIs, agent products or the infrastructure behind them. - Daily fluency with coding agents like Claude Code, Codex, OpenCode, Pi or Hermes, and specific opinions about what makes an agent loop, a harness or a tool integration good. - A solid grasp of how agents work under the hood: tool calling, context management, prompt caching, evals, and the cost, latency and quality tradeoffs between models. - Good systems judgment for running agents safely: sandboxing, process isolation, secrets, and reliability across many users. - High agency and a bias toward action. You talk to users, prototype, ship and iterate on a small team with fluid boundaries. - Product taste for developer tools. You notice interaction details and care how the first five minutes feel. NICE TO HAVE - Built or maintained an open-source agent harness, CLI or developer tool. Experience with multi-agent systems, agent skills or memory, or eval and benchmarking tools.Shipped a Slack app or a native macOS app.Existing user of OpenRouter or Ori, or active side projects in AI tooling. If you don't think you meet all of the criteria above but are still interested in the job, please apply. Nobody checks every box, and we're looking for someone excited to join the team.

Requirements

Must-haves

Build and ship across every Ori surface: the ori CLI and its harness integrations.
Make model fungibility real in any harness. That means model switching, provider fallbacks, tool and prompt settings tuned to each model, and respecting org guardrails, all without extra auth for the user.
Take agents from a laptop to the cloud: deploy, fork and shut down agents on command, with agent state that is version-controlled and auditable.
Own quality end to end, from install, auth and CLI UX to reliability, observability and evals. Ori is heavily dogfooded internally, so you'll use what you build every day.
Feed the gaps Ori exposes back into OpenRouter's API and SDKs, so the whole ecosystem can build multi-model agents on the same primitives.
5+ years shipping production software, with real ownership of developer tools, CLIs, agent products or the infrastructure behind them.
Daily fluency with coding agents like Claude Code, Codex, OpenCode, Pi or Hermes, and specific opinions about what makes an agent loop, a harness or a tool integration good.
A solid grasp of how agents work under the hood: tool calling, context management, prompt caching, evals, and the cost, latency and quality tradeoffs between models.
Good systems judgment for running agents safely: sandboxing, process isolation, secrets, and reliability across many users.
High agency and a bias toward action. You talk to users, prototype, ship and iterate on a small team with fluid boundaries.
Product taste for developer tools. You notice interaction details and care how the first five minutes feel.

Nice-to-haves

If you don't think you meet all of the criteria above but are still interested in the job, please apply. Nobody checks every box, and we're looking for someone excited to join the team.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aiapi

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Pto