Medjai
Medjai provides a complementary decision layer for AI agents, capturing human overrides and converting them into versioned, validated rules that agents follow across any stack. The platform uses an SDK and MCP integration to self-report corrections, cluster them, and require human validation before publishing rules with full provenance. It continuously measures adherence through repeat-override tracking, creating an auditable system of record for regulated operations.
- Artificial Intelligence
- Regulatory & Compliance Technology
- AI Agents
- Enterprise Software
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
As AI agents scale across operations, the human corrections and overrides they receive represent the most valuable judgment data a company produces, yet this knowledge typically vanishes. Organizations waste scarce expert time on repeated corrections, agents fail to learn from past mistakes, and institutional expertise walks out the door when experienced employees leave.
Solution
Medjai sits below an organization's agents, not in place of them, capturing every human override at the source through its SDK and MCP integration. The platform clusters these overrides, drafts candidate rules, and requires human validation from company experts before publication. Once published, versioned rules deliver to every subscribed agent across any stack, with adherence measured continuously through the system's loop—if a rule is not followed, the override returns and compounds the corpus. This creates a durable, auditable system of record where agents inherit years of judgment on day one and every decision traces back to who set it, when, and under what authority. The company frames its role as "the complementary decision layer" that collapses the space of possible agent actions around what matters, a need that becomes more acute as physical AI expands action possibilities.
Target Audience
Primary customers are regulated operations running decision agents on platforms like Microsoft Copilot, Salesforce Agentforce, or custom enterprise stacks—including RevOps teams, insurance underwriters, and financial services firms that need demonstrable governance over AI decision-making. The company also positions its platform for organizations deploying physical AI systems where actions carry real-world consequences.
Features
- Agent self-reporting of overrides via Medjai SDK and MCP, eliminating log-scraping and guesswork about what counts as a correction
- Automated clustering of overrides with candidate rule drafting, requiring explicit human validation before any rule ships
- Versioned rules with full provenance, capturing the source, author, context, and authority behind every decision
- Cross-platform rule publication to all subscribed agents regardless of stack, with rule sets surviving vendor switches
- Built-in compliance measurement through repeat-override rate tracking, providing a demonstrable oversight metric for auditors and regulators
- Live corpus that compounds over time, functioning as institutional memory rather than static documentation