AI Agent PlatformWork shows up.
CatalEx moves.

CatalEx is the AI operating layer for AI-native companies — one platform to build, deploy, and run AI agents. Agents can be embedded in your product via API, work autonomously in the background on scheduled or long-running tasks, or sit beside your team as named AI assistants in tools like Slack, Notion, Jira, and GitHub.

Describe the work in plain words — like briefing a new hire — and CatalEx assembles a team of specialist agents and wires the handoffs. Agents handle work such as lead research and qualification, personalized outreach, meeting scheduling, invoice reconciliation, cohort reports, onboarding verification, claims processing, and compliance checks.

CatalEx agents are proactive, measured, and evolving rather than reactive. They carry persistent memory across sessions, catch and fix their own mistakes mid-run, pick and track their own business metrics, and learn from every outcome — so performance improves over time instead of staying the same as day one.

CatalEx has a Free plan ($0) with 2 agents, 500 credits per month, all integrations, and schedules — no credit card required. The Pro plan is $20 per month with 5 agents, 2 users, and 5,000 credits per month. Enterprise pricing is custom and includes SSO, on-premises deployment, audit exports, and forward-deployed engineering.

Yes. CatalEx offers role-based access control inherited from your identity provider, a full audit trail of every decision and tool call, human-in-the-loop approval workflows, and on-premises or in-region deployment so data never leaves your perimeter.

CatalEx is model-agnostic — it works across Claude, GPT, Gemini, and Llama with no vendor lock-in. Intelligent model routing, caching, and code re-use make runs up to 10x cheaper.

An AI agent platform is the software layer where you build, deploy, run, and measure AI agents — autonomous programs that use a language model to plan work, call tools, and complete multi-step tasks without a person driving each step. A platform supplies what a raw model API does not: tool and integration plumbing, persistent memory, scheduling, permissions and approvals, an audit trail, and measurement of whether the agent actually did its job. CatalEx provides these as one runtime rather than a stack you assemble yourself — see the documentation for how the pieces fit.

Yes, within limits you set. CatalEx agents run on a schedule or trigger and work in the background on long-running tasks like invoice reconciliation, cohort reports, and lead outreach. Full autonomy is rarely the goal, though: agents that touch money, customers, or production should pause for a human. CatalEx handles this with human-in-the-loop approval steps, so an agent runs unattended for routine work and stops for a sign-off on anything sensitive — with every decision and tool call recorded in the audit trail.

Tool use is how an agent acts on the world instead of only describing it. The model is given a set of tools — each with a name, a description, and a schema for its arguments — and it decides which to call, with what arguments, and what to do with the result, looping until the task is done. CatalEx exposes roughly 100 integrations as tools, supports custom tools and MCP servers, and scopes every tool to the permissions you grant, so an agent can only reach what its role allows. We wrote up when to reach for MCP versus a plain CLI.

Agentic development is building software where an AI agent — not a person — performs the steps: you specify the intended outcome and the checks that prove it, and the agent plans, executes, and revises against them. It shifts the work from writing procedures to writing specifications and evaluations. The practical consequence is that the harness around the model — its tools, memory, context, and evals — matters as much as the model itself, which is why CatalEx treats the runtime and the eval loop as the product rather than as glue code.

With evals, not vibes. Define the outcome the agent is responsible for, then score every run against it. CatalEx agents pick and track their own KPIs, score themselves after each run, and record what worked and what needs improvement, so you see trends over time rather than a single impressive demo. This is the difference between an agent that looks good on day one and one that is measurably better on day 100 — and it is the reason most AI pilots stall before production.