What Are MindsHub Agents, Code, and Inference Used For?

MindsHub splits its work-facing products into three surfaces: Agents for delegated research and app-building, Code for repository work like PR review and bug fixes, and Inference for calling models from your own code. They sit on one platform with one model catalog, so the choice is about the shape of the task, not about which vendor you're locked into.

Agents: delegate a task and get a report or app back

Agents is the workspace for tasks you'd otherwise hand to a person: research, analysis, and building something presentable.

The site's own example is a KPI app. A user asks for an interactive app the team can explore — MRR, signups, and NRR pulled live from Postgres and Stripe, with the ability to open a Linear ticket from a flagged row and publish a shareable URL. The agent then:

• Plans the analysis (3 KPIs, live refresh, drill-down by segment)
• Connects to PostgreSQL through a credentials vault, so no raw key is exposed
• Queries Stripe for MRR and signups, reconciling 1,204 subscriptions
• Wires up Linear so issues can be created straight from flagged rows
• Publishes kpi-app.html as a live, shareable URL

The result reported back: MRR up 9.4% week-over-week to $128.4k, plus a flagged finding that signup checkout was erroring for EU accounts.

Use Agents when the deliverable is a document, a report, or a small interactive app, and when the task needs connectors to outside systems. The workspace also lists Projects, Scheduled Tasks, Live Artifacts, Connected Apps and Data, Agent Memories, and a Skills library — so recurring work (a daily 7am briefing across Slack, for instance) and reusable skills are part of the same surface.

Code: work inside your own repositories

Code is the surface for engineering tasks against repos you already own. The listed task history shows the pattern:

The description is explicit: review PRs, fix bugs, and ship features in your own repos. Use Code when the output is a change to a codebase rather than a report — and note that the tasks above are scoped and reviewable, which is the useful unit of delegation here.

Inference: one key, one bill, many models

Inference is the developer-facing piece: plug models into your code and call major models with one key and one bill. The site names Kimi K3, GPT-6 Astra, and Claude as examples. The positioning is a drop-in replacement for OpenAI and Anthropic, with the same SDKs, apps, and capabilities you're used to.

Use Inference when you're building AI into a product rather than using it to get work done. The value proposition stated is model choice plus control over cost and governance — not a single model.

How the three relate

All three draw on the same model selection: the platform's stated purpose is giving everyone the freedom to choose among leading AI models and providers, whether they use AI to get work done or build it into products. Practically:

• Agents — you describe an outcome; the platform plans, connects to data, and produces an artifact.
• Code — you point at a repo and a specific engineering task.
• Inference — you write the code and call models through an API.

A task can cross surfaces: the KPI app example used connectors and produced a published artifact, which is Agents territory, while the same platform's Code surface handled PR review and bug fixes on a separate repo.

Access and cost

MindsHub Inference is described as free to use with the MindsHub Air model, and the site states no credit card is required, with a linked free-tier page. The pricing page is at mindshub.ai/pricing. Beyond the Air model's free tier, the source material does not state specific prices or plan limits — check the pricing page for current terms rather than assuming the whole platform is free.

There is also a guest Playground and a desktop app listed in the interface, which is a low-commitment way to try the model selection before wiring anything up.