Bring the model. Melaya makes it an agent.
A model on its own can reason. It cannot act, remember, wait for a schedule or ask before it sends. Melaya Agents turns whichever model you pick, hosted, routed, local or private, into an agentic system with tools, memory, schedules, approvals and replayable runs.
- Scoped tools and connected apps
- Memory across runs
- Schedules and triggers
- Approvals before writes
- Replayable runs
Before an AI agent enters a real workflow.
Direct answers about the Melaya AI agent builder: free and no-code use, integrations, model choice, permissions, deployment, and operational control.
Is there a free AI agent builder plan?
Yes. The Sandbox plan is free and needs no credit card. You build pipelines on the same canvas and run them in Melaya's cloud with your own AI key, or on your own computer through the Melaya runner: 10 pipeline runs in total, cloud or local. Outpost adds 50 runs a month, 30 of them in the cloud, and Forge adds longer cloud runs and schedules.
Can I build AI agents without code?
Yes. Start from a ready-made template, describe the workflow and let AI-assisted creation draft it, or drag personas onto the canvas and brief each one in plain English. Prompts, schemas, models, and tools stay editable for teams that want finer control.
Is Melaya an enterprise AI agent builder?
Yes. Each agent gets only the tools it is granted, consequential writes can wait for human approval, and every run is traced and replayable. Bastion adds teams and shared projects, and Citadel adds enterprise deployment, SSO, an SLA, and custom capacity.
What is a visual AI agent builder?
A visual AI agent builder turns a business process into an executable graph of agents, models, tools, context, triggers, and approval gates. In Melaya, teams can inspect that graph, run it, schedule it, and replay every step without hiding the operating logic inside one prompt.
Can Melaya build enterprise agents for Odoo, Stripe, Google, Zoho, Slack, and other systems?
Yes. A workflow receives only the connectors and tool actions explicitly granted to it. Teams can combine business systems in one pipeline, correlate their results, and require approval before a write such as sending, publishing, updating a record, or making a payment.
Do we need to code every AI agent from scratch?
No. Start from a ready-made pipeline template, describe the workflow to AI-assisted creation, or compose it directly on the visual canvas. Technical teams can still refine prompts, schemas, models, tools, memory, and execution policy at each step.
Can we use our own AI models or run models on enterprise infrastructure?
Yes. Choose a hosted provider with your own key, a router such as OpenRouter or LiteLLM, a private OpenAI-compatible endpoint (vLLM, llama.cpp server, a company gateway), your Claude Code or Codex subscription, or local models through Ollama and LM Studio. Model choice can vary by agent, so private steps can stay local while other steps use a cloud model.
How do permissions and credentials work in the agent builder?
The builder grants tools explicitly. Ungranted tools do not enter the agent's available schema, connector credentials remain server-side, and tenant-scoped access limits which projects, data, devices, and actions a run can reach.
Can a human approve an agent action before it happens?
Yes. Human-in-the-loop gates can pause consequential actions and show the proposed arguments for review. The operator can approve, edit, or reject the action, while the decision remains attached to the run trace.
How do teams test, monitor, and audit AI agent workflows?
Every run records its steps, model rounds, tool calls, results, approvals, costs, and terminal outcome. Teams can compare runs, replay prior executions, inspect failures, and use evaluation evidence instead of trusting the agent's final message alone.
Can Melaya be deployed or embedded for enterprise clients?
Yes. Melaya supports cloud, local-runner, and enterprise-server operating models, plus SDK-based integration and white-label delivery for qualified implementations. Deployment, model routing, capacity, and support can evolve with the client's security and usage requirements.
Anyone can build agents. Build your first pipeline.
Drag personas onto a canvas, wire the steps, and ship a governed pipeline with HITL gates, shared memory, and full run replay built in.
Free to start · Human control built in · No model lock-in