Mrrlin is a transparent AI execution workspace for founder-led teams. Its positioning is not a chatbot, nor a task board, but a command layer that can truly "get work done": users only need to state a business goal in one sentence, and the system will complete the full closed loop from intent clarification to final delivery.
Purpose
Mrrlin is designed to help small teams avoid becoming "AI project managers." Typical uses include: launching marketing websites and SEO foundations, building the first growth loop; finding target customers and preparing personalized outreach campaigns; turning customer feedback into scheduled product specs; auditing client projects and outputting improvement plans; creating a planning package for marketing campaigns that includes positioning and competitive research.
Main capabilities
- Goal to execution plan: A natural-language business goal is transformed into scope, constraints, acceptance checks, and visible tasks.
- Upfront clarification: If context is missing, Mrrlin proactively asks questions before the agent starts working.
- Agent routing: Routine work is assigned to cheaper models; difficult or code-heavy tasks are routed to Claude Code, Codex CLI, Sonnet, DeepSeek, or local agents.
- Consensus review: Important decisions undergo independent model review, iterating over multiple rounds until convergence, checking evidence, assumptions, brand fit, and acceptance criteria.
- Weak output rejection: Rejected drafts, weak research, or unsound implementations are sent back for rework with review comments.
- Evidence gate: Every task must provide verifiable proof (a usable address, a green build, or user sign-off); a model claiming "done" is not considered evidence.
- Human approval inbox: Sensitive claims, sends, publications, deployments, and final deliveries wait for human approval.
- Project memory: State, decisions, and evidence are stored in the Mrrlin database and persist across sessions.
- Isolated runtime environment: Each run takes place in its own git worktree, and artifacts are recorded throughout.
How to use
Users provide a goal, and Mrrlin plans and then dispatches tasks to agents. Routine work continues automatically, and weak output is automatically sent back; when publication, sending, or client-visible claims are involved, it pauses and waits for user approval. Models run on the user's own machine, using the user's own keys, subscriptions, and models; Mrrlin does not hold model credentials and does not resell tokens. The system is designed to work across existing tools and does not force migration.New system. The official page mentions no credit card required, no setup required, and starting from one goal.
Applicable scenarios
- Founder-led SaaS: produce sharper marketing websites, SEO foundations, analytics events, and the first growth campaign without personally managing AI.
- AI-native agencies: turn client briefs into campaign strategy, content, landing page variants, and final QA.
- Growth/product teams: run acquisition and onboarding experiments end to end, keeping approvals and metrics visible.
- Technical agencies: turn client requests or repository audits into scoped work, fixes, tests, and reports.
Limitations
According to the available materials, Mrrlin's pricing, specific integration methods, supported operating systems, agent configuration details, data processing policies, and other details are not yet provided. It does not promise to replace users' existing tools, but instead works across existing tools. Humans retain review authority over final delivery and sensitive decisions.
