xTrac Agent Builder -with Real Memory and 100+ Integrations

AI-native developer platform to build multi-agent workflows with real memory, integrated schema-driven forms, HITL approvals, and 100+ secure app integrations. Open AI Competitor for Agent Builder.

Comment

The rise of AI agents unlocked new possibilities, but every tool we evaluated be it OpenAI's early Assistants, CrewAI, Zapier or n8n - either lacked memory, structured forms, or human-in-the-loop (HITL) safeguards. Overall, it needed another tech team to build the agents or agentic workflows which are still static agents.

We realized there was no platform that combined multi-agent workflows that can fan out agents during runtime without being static, real-time HITL review, and integration with 100+ SaaS tools all orchestrated with reliable stateful memory, first of its kind in the agentic world. And all this can be done with just one prompt.

This inspired us to build xTrac Agent Builder with AI-first orchestration layer that empowers organizations to design and deploy dynamic, agentic workflows combining AI tools, structured forms, business logic, and real humans all using natural language.

How We Built It

We built xTrac from the ground up as a full-stack system:

Backend: Python + Django with PostgreSQL

Inbuild State Management

Inhouse 100+ secure OAuth app integrations

Modular AI Nodes: Agent, HITL, Integration, Decision, Fallback, Form

Frontend: React 19 + Next.js 15 with TailwindCSS

AI Prompt or Canvas for building workflows

AI-generated schema-driven forms with validation

Test Playground with step-by-step execution

Core Features: 100+ SaaS integrations via Nango Real-time AI memory with checkpointing Human In The Loop reviews with audit logging Modular & Stateful execution engine AI-powered form generation & editing

What We Learned- Multi-agent systems need memory: Stateless execution leads to hallucinations. So we implemented real memory with persistent checkpoints and DB logging.

HITL is not optional: For real-world adoption, workflows need human approvals, fallback states, and audit trails.

Drag-and-drop isn't enough: We needed full test environments, prompt observability, and fallback recovery to make workflows usable by real teams.

Integration is everything: Our users wanted Gmail, Zoho, Salesforce, Slack, Google Sheets, etc., out of the box. Developer friendly OAuth that saves months of work.

Challenges We Faced Building Workflow and Stateful execution Engine: Building composable multi-agents that support fallback, retry, and dynamic injection without breaking the DAG was hard.

Secure RBAC + PII masking: Enterprise teams needed field-level privacy with full RBAC, which we built into every node and UI component.

Editable schema-driven forms: We built an AI-first form renderer with schema editing, RBAC, and preview-safe rendering.

Testing agent workflows in real time: We created a live Test Playground that supports step-by-step or auto-run of any AI workflow with form injection, agent tracing, and integration mocks.

Vision We're just getting started. Our long-term vision is to build the Kubernetes of AI Agents- a platform where teams can design, deploy, and monitor self-healing, multi-agent, HITL-secure AI systems across any domain.

Share this project:

Updates

Submission history