MiroFish

MiroFish spawns thousands of AI agents, each carrying its own memory, beliefs, and influence profile, and sets them loose across a simulated dual-platform social environment that mirrors real-world dynamics like Twitter and Reddit.

Agents argue, shift opinions, form coalitions, and exhibit emergent herd behaviour. This gives strategists a far more accurate picture of public or stakeholder reaction than any single-model output ever could, capturing polarisation and coalition shifts before they happen in reality.

Rather than working from generic training data, MiroFish reads your uploaded documents, PDFs, Markdown files, or plain text up to 30 MB, and converts the content into a structured knowledge graph. This graph defines the actors, incentives, and constraints that shape agent behaviour during the simulation.

The practical benefit is that every prediction report is grounded in your actual business context, not a statistical average of someone else's data.

Once a simulation is running, users can inject new variables from a top-down control perspective to see how the agent population responds to a changed input. This is the feature that separates MiroFish from statisk analyse tools. You can stress-test multiple versions of a product launch message, a pricing change announcement, or a policy draft within the same session and compare emergent outcomes across each scenario.

Every completed simulation outputs a fully structured report containing an executive summary, identified risk warnings, stakeholder reaction narratives, and quantified confidence scores. Reports are exportable as clean, board-ready PDFs in one click. This is particularly valuable for communications teams and executives who need to present scenario analysis to a board without rebuilding the data from scratch.

After the simulation concludes, users can directly message individual agents to interrogate their reasoning, test edge cases, or explore minority viewpoints that the headline report may not have surfaced. This is a meaningful capability for markedsforskere who want to dig into outlier behaviour or product managers who need to understand the specific segment driving a risk signal.

MiroFish ships a full Docker Compose deployment option with GHCR image hosting for one-click local setup. Organisations with strict data governance requirements can run the entire stack on their own infrastructure, supplying their own LLM API keys and Zep Cloud configuration for memory management. For enterprise teams in regulated industries, this self-hosted path provides full data control that SaaS-only competitors cannot match.