A model-agnostic alternative to OpenAI dots

September 29, 2026 - 4 minute read

OpenAI launched dots on September 29, 2026, bringing ongoing work, connected apps, and cloud computers into one agent experience. For software teams that want this kind of delegation with a choice of models, Factory offers a model-agnostic approach.

A team may need an approved provider for one repository and a locally hosted model for another. Factory's model options let those requirements guide the setup, while its engineering workflows organize planning, implementation, and validation.

OpenAI dots and Factory on model selection

OpenAI dots. Dots uses GPT-6 Astra at launch. OpenAI's announcement does not describe a bring-your-own-model option.

Factory. Custom Models support provider keys, compatible model endpoints, and locally hosted models in the Droid CLI and desktop app. Those custom configurations do not appear in the hosted web or mobile products. API compatibility is necessary, but it does not guarantee that every model works out of the box.

Factory Router adds automatic model selection based on quality, latency, cost, and cache state. Choose Custom Models when you need to configure a particular endpoint. For routed selection, check that Router's supported configuration meets your provider requirements.

Factory is worth evaluating when you need a particular provider or want routing rather than one fixed model choice. Test that setup with the repository's tools and validation procedure. A model that writes a plausible patch but cannot complete the checks may not suit the workflow. Confirm the interface you will use instead of assuming a CLI configuration applies everywhere.

OpenAI dots and Factory pricing

OpenAI dots. The first dot is included in eligible Pro and Business Premium plans, with an allowance for deeper work. OpenAI describes expanded limits during the first month. Tasks delegated to Codex or ChatGPT Work retain their own usage limits. The plan's allowance therefore matters alongside access to the dot itself.

Factory. Individual subscriptions are Pro at $20 per month, Plus at $100, and Max at $200. They have independent rolling 5-hour, 7-day, and 30-day usage limits. Managed Droid Computers are included from Plus. Extra Usage is prepaid credit, and purchased credit carries over while unused standard usage does not.

Factory's organization plans separate self-serve Teams from custom Business and Enterprise agreements. Teams costs $60 per month plus $40 per seat, for up to ten seats with Pro limits per seat. Business and Enterprise have custom usage limits and contract terms.

Estimate the bill using the subscription and the usage needed for your expected workload. Run representative tasks to see how quickly they consume each service's allowance, and account for the end of any introductory limits. This is especially useful for recurring jobs whose usage accumulates throughout the month.

Cloud computers and machine ownership

OpenAI dots. Dots have their own cloud computers and browsers. You can inspect a dot's computer while it works and approve connections to other devices, including your laptop.

Factory. Droid Computers retain files, packages, services, and configuration across sessions. Factory can provision a managed computer, which can pause when idle and resume for new sessions. Alternatively, you can register a machine your team controls.

The BYOM guide explicitly includes cloud VMs, VPSs, workstations, and on-premises servers. This is the Factory alternative when you want work to run in an existing engineering environment. Your team remains responsible for its hardening and maintenance, and the machine needs network access to Factory's services and relay.

BYOM requires network access to Factory, so it is not an airgapped configuration. Include credentials, dependencies, and persistent session state in the machine's security review.

Project planning and execution

OpenAI dots. A dot can carry several projects forward and preserve context across conversations. The launch includes building and testing software improvements, returning pull requests, and attaching videos for review. It also covers non-engineering work such as research and proposal preparation.

Factory. Missions support bounded software projects through collaborative planning, milestones, worker coordination, and validation. You approve the plan before execution. The task's success criteria are part of the project rather than something a reviewer has to infer from the final patch.

For a multi-feature project, specify the intended behavior before work starts and review the implemented scope, checks performed, and unresolved issues afterward. Missions gives that engineering work a planned structure. Dots also supports projects such as research and proposal preparation that extend beyond software delivery.

Starting work in the background

OpenAI dots. Proactive research uses read-only connected-app tools to look for useful work between conversations. That discovery mode cannot send messages, change app content, or control a computer. Finding a possible task and acting on it have separate permission boundaries.

Factory. Custom Automations start sessions from a schedule or a configured Slack, GitHub, or webhook trigger. Webhooks remain in private preview. The instructions and run configuration define the work that a matching event is allowed to start.

The distinction is how a background task begins. Dots can discover possible work from connected context. Factory starts from an event or cadence you configure, such as reviewing a particular class of incoming report. Define what should happen after that trigger and who handles the result. A daily check and a response to a new bug report may need different instructions.

Permissions and organizational controls

OpenAI dots. Users control app access and define Custom Rules that allow actions, require approval, or block them. Activity View exposes progress, and action auto-review checks proposed actions against instructions and safety requirements. Certain sensitive actions remain with the user.

Factory. Enterprise Controls let administrators define enforced boundaries around models and tools. A user preference does not override a hard organizational control. That distinction matters when model flexibility must operate within an approved provider policy.

Compare permissions against the same proposed task. A worker may need to read an issue, edit a repository, and run tests without having authority to merge the result. Check which actions are permitted, which require a person, and which are blocked. Model flexibility should operate within those boundaries rather than changing the authority granted to the agent.