Minicor – Self-Healing Desktop RPA at Scale

TL;DR

TL;DR: Minicor runs AI agents on Windows desktops that automate legacy software (EHRs, ERPs, DMS) without APIs — the agents self-heal when UIs change, achieving 93-96% click accuracy.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: minicor.com
• Documentation: docs.laminar.run
• GitHub (organization): github.com/laminar-run
• License: Proprietary (enterprise SaaS)
• HN launch thread: news.ycombinator.com/item?id=48445171
• Source last checked: 2026-07-22

What Is Minicor?

Legacy desktop software — healthcare EHRs, dental PMS, automotive DMS, supply chain WMS — has no APIs. The only way to read or write data is to interact with the desktop like a human would. Minicor automates these systems reliably at scale using computer use agents that run on Windows desktops or VMs.

The platform stores automations as deterministic code for speed, then uses a reasoning model, grounding model, reflection agent, and OCR only for recovery and adaptation. When a UI element moves or changes, the agent detects the failure, re-scans the screen, and self-heals — the workflow keeps running instead of breaking.

Key differentiators from the README:

• 93-96% click accuracy versus 80-85% for generic computer use approaches
• Self-healing: agents adapt when UIs change instead of failing
• One API call triggers a full desktop workflow and returns structured JSON
• Video replay of every action with error logging and monitoring
• Deploy on-premise, cloud, or Citrix — not locked to one environment
• SOC 2 Type II and HIPAA compliant
• Y Combinator W26 batch

Minicor also surfaces an MCP server endpoint, so AI coding agents can build, debug, and manage RPAs conversationally through a standard Model Context Protocol interface.

Setup Workflow

Step 1: Install the Desktop Client

Minicor runs a lightweight desktop client on the Windows machines where your legacy software lives. The client communicates with Minicor’s cloud orchestration layer.

Note: Detailed installation steps require a demo booking. The product is marketed as “zero to production in weeks” with guided onboarding.

Step 2: Record a Workflow

You record a video of a human performing the target workflow once. Minicor extracts the deterministic automation steps from the recording and stores them as structured code.

Step 3: Trigger via API

One API call kicks off the full desktop workflow on the target machine. The agent executes steps autonomously, self-healing on any UI mismatch:

curl -X POST https://api.laminar.run/workflow/execute \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "configuration_id": "30",
    "machine_id": "windows-vm-prod-01"
  }'

The response is structured JSON with the workflow output.

Step 4: Monitor via Video Replay

Every action is recorded. You can replay any run, inspect error logs, and see exactly where self-healing triggered.

Target Systems

Minicor publishes native integrations for:

• Healthcare EMRs: Athena, eClinicalWorks, Epic, Cerner, PS Suite, Meditech
• Dental PMS: Open Dental, Dental Vision, Eaglesoft
• Automotive DMS: CDK Global, Reynolds, Techeon
• Supply Chain WMS/ERP: HighJump, SAP, Sage 300
• Home Health: Home Care HomeBase, Wellsky, Kinser
• Financial Services: claims platforms, banking, underwriting

Minicor vs Alternatives

| Approach | Self-Healing | API-First | On-Prem | Video Replay | |---|---|---|---|---| | Minicor | Yes | Yes | Yes | Yes | | UiPath/Blue Prism | No | Partial | Yes | No | | Anthropic/OpenAI Computer Use | Partial | No | No | No | | Internal RPA build | Per-case | Varies | Yes | No |

Minicor’s architecture stores automation as deterministic code — the agent reasoning layer only kicks in for recovery, which is why its production accuracy (93-96%) beats general-purpose computer use models that rely on pure model inference for every action.

Security and Compliance

• SOC 2 Type II certified ( audited via app.mycroft.io/trust/laminar )
• HIPAA compliant — relevant for healthcare automation
• Runs inside your own Windows environment (on-prem, cloud VM, or Citrix) — data does not leave your infrastructure without explicit configuration

FAQ

Q: Does this work with cloud-based software or only thick-client Windows apps? A: Minicor targets Windows desktop applications. Cloud apps with browser interfaces can be automated through browser-based approaches; the README specifically calls out thick-client legacy systems as the primary use case.

Q: How does self-healing work in practice? A: When a UI element the agent expects to find is missing or has moved, the reflection agent triggers a re-scan using OCR and visual grounding. It re-locates the target and resumes the workflow rather than failing.

Q: Is there a self-hosted option? A: Yes — Minicor can be deployed on-premise or in your own cloud environment (AWS, Azure, etc.), not just as a managed SaaS product.

Q: What Windows versions are supported? A: The desktop client targets modern Windows environments. Specific version requirements are available in the docs.laminar.run.

Q: How is pricing structured? A: Minicor is enterprise SaaS. Pricing is not publicly listed — contact the team for a demo and quote.

Conclusion

Minicor solves a specific and painful problem: automating legacy Windows desktop software that was never designed for programmatic access. Its self-healing architecture (93-96% click accuracy) and compliance posture (SOC 2, HIPAA) make it a practical option for AI-powered automation in regulated industries like healthcare, automotive, and logistics. If you are running a business that depends on thick-client software with no API, Minicor is worth a demo.

• Product page: minicor.com
• Documentation: docs.laminar.run
• Book a demo: minicor.com/book-demo