Bring more coding agents into OpenCircle
Coding harnessesAgent protocolsDeveloper toolsTesting
A couple of terminal coding agents already run inside OpenCircle. Help us connect the ones people ask for next and give them the same seat at the table: your Familiar hands the job over, watches it progress, brings the result back, and you stay in charge the whole time.
A good first step
Choose one agent that isn't supported and extend the existing harness integration. Carry it through a whole job — connect, pick a workspace, start, handle an approval, redirect, stop and resume where the tool allows. Keep accounts apart, keep diffs easy to review and make failures loud. Extend the handoff path we have instead of inventing a second one.
Use the AI subscriptions you already pay for
Provider integrationsOAuthSubscriptionsSecurity
Help people plug in the chat subscriptions they already have through each provider's supported sign-in route. Make it obvious which account is linked, what it may do and when work draws on a subscription instead of metered API usage. Nobody should have to guess which bill is growing.
A good first step
Prove one provider's official path end to end, from sign-in to useful work, then test expiry, renewal, sign-out and running out of quota. Keep credentials protected, respect account boundaries and spell out plan limits before anyone begins.
Connect the services people really use
IntegrationsBrowser UseAPI designSecurity
Grow the connector-based integration path, or push the automated browser through repeatable tests on real websites. We need connections that survive expired sessions, interrupted jobs, changed permissions and the second attempt.
A good first step
Pick one service and one ordinary task. Make sign-in, the task and recovery all work end to end, with tests and plain setup notes. One successful run is a beginning — not proof that a whole site is supported.
Take OpenCircle to more infrastructure
DevOpsInfrastructureKubernetesPostgreSQL
Docker Compose and Railway are where we started. Help write deployment drivers for Kubernetes, the big clouds and the other environments operators already run, with installs, upgrades and recovery that feel routine.
A good first step
Walk one target through deploy, upgrade, a deliberately interrupted upgrade, restore and teardown. Show what persists, what it costs and what breaks. Broader deployment choice should rest on real operating experience.
Make the phone experience even better
MobileFrontendUX/UIAccessibility
Mobile is a first-class way to use OpenCircle on iPhone and Android, and we want it sharper. We're looking for frontend engineers with a designer's eye for touch targets, motion, spacing, keyboards and the tiny details that make daily use effortless.
A good first step
Take one flow onto a real phone. Find the pause, the cramped control, the keyboard covering the next step. Bring a small before-and-after prototype and prove it with accessibility settings on, a weak connection and the app waking from the background.
Polish the desktop details
FrontendUX/UIElectronAccessibility
OpenCircle has a clear design direction and we want builders who enjoy bringing it to life: rich editors, crisp keyboard navigation, panels that resize well, and screens that stay calm while several things happen at once. Opinions welcome; committees that sand off the personality, less so.
A good first step
Find an interaction that's nearly right and make it right. Record the difference and check it across window sizes, input methods and both light and dark themes.
Make Windows a great home for OpenCircle
WindowsElectronFrontendTesting
The desktop app is built on Electron. We need Windows developers and testers who know where cross-platform promises crack: installers, updates, permissions, file paths, audio devices, display scaling and the habits Windows users expect.
A good first step
Run an everyday workflow on Windows, write down what breaks and bring a focused fix with a test that reproduces it. We want confidence in the whole experience, not just a window that opens.
Give Familiars a cloud desktop
Cloud desktopsInfrastructureStreamingSecurity
We're building cloud computers that Familiars can operate while a person watches and can take over. There's room for people who understand remote desktops, streaming, isolation, long-lived sessions and keeping a machine available and recoverable.
A good first step
Help prove one full session: boot the machine, do real work, hand control to a person, reconnect after a drop and shut down cleanly — with cost and boundaries made clear.
Let agents pay, with permission
PaymentsSecurityAPI designTesting
Help design responsible agent payments on top of established payment infrastructure. This is real money: we need people who think about spending authority, repeated charges, hijacked sessions, refunds and reconciliation before celebrating a checkout.
A good first step
Propose one narrow purchase flow in a provider sandbox. Show exactly what the person approved, how spending is capped, how credentials stay out of model context and what happens if the connection drops before the receipt. Never guess that a payment failed and charge again.
Build mini-apps for people and Familiars
Mini-appsFrontendFamiliar toolsTesting
Build something you'd use every day. A good OpenCircle mini-app gives a person direct, satisfying controls and gives a Familiar precise tools for the same object, so neither has to wrestle the other for the document, design, sum or plan.
A good first step
Bring a small idea and a working interaction: a person editing, a Familiar making a change, the person taking over again. Test saving, undo, interrupted edits, permissions and reopening until it's dependable.
Find the holes before someone else does
SecurityThreat modelingTestingAgent safety
We want researchers who enjoy finding the assumption everyone else stepped over. Probe permissions, cross-user data access, tool execution, dependencies and the ways hostile content can nudge an agent into doing something it shouldn't.
A good first step
Only test systems you own or are explicitly allowed to assess. Bring the smallest case that reproduces the issue and, where you can, a regression test. Report suspected vulnerabilities privately — never in a public issue or pull request.
Run models on hardware you own
Local AIModel runtimesInfrastructure
Help explore local inference that makes real work possible on your own machines. We want people who like measuring what models and runtimes can actually do — including where the experience starts to fall apart.
A good first step
Bring one measured model-and-hardware experiment with setup steps, representative tasks and honest limits. Build on the existing provider architecture rather than a separate agent stack.
Federation: link independent circles
Distributed systemsProtocol designSecurity
Help research how separately run OpenCircle servers could work together while each keeps its own members, keys and rules. Start from the people who need to collaborate and what they should be able to share.
A good first step
Bring a concrete collaboration story and a small protocol experiment. Explain identity, consent, removal and what happens when one server goes offline, so we learn before committing to a big build.
