MINA LABS
Start creating free
OpenAI has introduced Presence, a platform for deploying voice and chat agents inside enterprise workflows. The system is designed to carry out repeatable interactions while maintaining consistent boundaries and verified outputs. Details are available in the full announcement at the primary source.
The offering focuses on reliability rather than novelty. Agents are built to follow instructions, surface source references, and decline requests that fall outside policy. Voice channels are treated as first-class interfaces, with latency and interruption handling tuned for real conversations instead of demonstrations. Internal tools can route tasks to agents without rewriting core logic, and logs capture decisions in formats that existing audit pipelines can ingest.
For teams that ship products, this shifts how interfaces are assembled. Designers can wire chat or voice directly into flows that previously required forms or phone trees, and engineers can enforce constraints at the agent level instead of scattering checks across frontends. The platform handles turn management and context windows, so builders can concentrate on outcomes rather than conversational mechanics. Support surfaces can be updated by adjusting instructions and retrieval sources instead of maintaining brittle decision trees.
Reliability is the practical benefit. Presence is positioned as a way to reduce variance in how questions are answered and tasks are completed. When a customer asks for a status or a refund, the agent follows the same sequence each time, using verified data and returning consistent language. Internal workflows gain similar stability, with agents checking permissions and formats before acting. This matters for teams that need predictable throughput and measurable error rates instead of occasional brilliance.
What it costs and whether it is on Mina Labs is the detail that sets this apart. Presence is available on Mina Labs today, and usage is priced at 8 per image generation request within workflows that include visual outputs. There is no setup fee, and billing aligns with actual agent actions rather than seat counts. Enterprises can test in staging without commitment and move to production with the same pricing surface.
We would use this to replace brittle phone menus and chat forms that require constant upkeep. Customer onboarding can be handled by voice agents that verify identity, explain steps, and record consent without forcing users into rigid scripts. Internal triage can route requests to the right team after checking permissions and priorities, then log the outcome for compliance. Documentation lookups can be done by chat agents that cite sources and avoid hallucinations, letting engineers stay in flow while getting accurate answers.
Longer term, the platform invites careful composition rather than wholesale replacement. A checkout flow might include a voice agent for accessibility, a chat agent for returns, and background agents that reconcile orders and update inventory. Each agent is constrained to a narrow scope, so failures are isolated and fixes are local. Builders can iterate on prompts, retrieval sources, and guardrails without touching core services, and roll back changes by reverting configuration rather than redeploying code.
The platform is not a replacement for human judgment or existing tooling. It is a way to give repeatable tasks to a system that follows explicit rules and logs its reasoning. For teams that need to move faster without sacrificing consistency, that is a practical trade. OpenAI Presence is available now, priced per action, and ready for integration into workflows that already value reliability over experimentation.