Mnemos Documentation

Mnemos is an AI intelligence layer built for Robinhood Chain. It unifies autonomous research, live market analysis, on-chain discovery, and persistent memory into a single console — so that understanding the chain becomes a conversation rather than a spreadsheet.

Where most tools show you raw data, Mnemos reasons over it. A team of specialist agents reads directly from on-chain and market sources, grounds every claim in a citable reference, and remembers what it learned so the next question starts from context instead of a blank page.

There is no wallet to connect and no sign-up wall. Open the console and start asking. A typical first session follows four steps:

1. 1

   Open the console

   Launch the workspace and you land directly in the intelligence console — the command surface for every capability.

2. 2

   Ask a grounded question

   Ask about a token, contract, market, or on-chain event. Mnemos routes the request to the right specialist agents.

3. 3

   Watch the reasoning

   Each run streams its real pipeline — understanding, consulting live sources, reasoning — so you can see how the answer is formed.

4. 4

   Build on the result

   Every answer is saved to memory and can be composed into an exportable report, so later questions build on earlier findings.

The platform is organized around a set of surfaces, each backed by the same grounded intelligence engine. You move between them freely; memory and context follow you.

A handful of primitives explain almost everything Mnemos does. Learn these and the rest of the product reads naturally.

Mnemos is a layered system. Each layer has a single responsibility and a clean contract with the next, which keeps the platform extensible as new data sources and models come online.

The intelligence engine turns a natural-language request into a grounded answer. It is deterministic in structure and transparent in operation — every stage is recorded and observable.

Prompt
  → Understand      (parse intent, select specialists)
  → Assemble        (recall memory + read live sources)
  → Reason          (specialist agents analyze in context)
  → Ground          (attach citations, flag limitations)
  → Compose         (synthesize a single answer + report)

Because each stage corresponds to a real recorded step, the reasoning shown in the console is not a decorative animation — it reflects work the engine actually performed, including which sources it consulted.

Mnemos runs a team of specialist agents rather than one general model. Each owns a domain and reports into the Mnemos Orchestrator, which coordinates them into a single grounded response.

Orchestration flows in one direction: the orchestrator decomposes the request, dispatches to the relevant specialists in parallel, and each specialist works from a shared context assembled from memory and live data. Their findings are reconciled — conflicts and gaps flagged — before a final answer is composed.

{
  "id": "chain-analyst",
  "role": "Network activity analysis",
  "capabilities": ["Read blocks and transactions over JSON-RPC"],
  "tools": ["chain.blocks", "chain.transactions", "chain.metrics"],
  "context": "orchestrator-assembled"
}

“Mnemos” is the Greek embodiment of memory, and persistence is the idea at the center of the product. Every finding, preference, and observation can be retained, so intelligence compounds instead of resetting with each question.

{
  "type": "research-finding",
  "scope": "long-term",
  "subject": "liquidity rotation",
  "importance": "high",
  "pinned": true
}

A research task is a deliberate, multi-source investigation. Rather than answering from a single lookup, the engine gathers evidence, reasons across it, and cites what it used — with an explicit note on anything it could not verify.

1. 1

   Scope

   The request is decomposed into the questions that must be answered to reach a conclusion.

2. 2

   Gather

   Specialists read live on-chain and market sources and recall relevant memory.

3. 3

   Synthesize

   Evidence is reconciled into findings, each tied back to a citable source.

4. 4

   Report

   A structured, exportable brief is composed and saved for future context.

Security and trust are foundational, not features bolted on later. The platform is designed to minimize what it holds and to be explicit about what it does.