Digital Life Agents

Multi-Agent Architecture: Academic digital life agents that work in parallel across disciplines.

AcaClaw ships five pre-configured academic agents — each a “digital life” character with its own persona, discipline expertise, Conda environment, and skill set. They run in parallel on the same OpenClaw gateway, each with an isolated workspace and session.

Table of Contents

Overview

Each AcaClaw agent is a fully isolated “digital life” with:

Property Description
Identity Unique name, emoji, persona, and behavioral guidelines
Workspace Isolated directory with its own files and memory
Environment Discipline-specific Conda env (bio, med, chem, phys, or general)
Skills Curated skill set matching the agent’s expertise
Session Independent chat history and session state

Agents share the same OpenClaw gateway but operate independently — you can chat with multiple agents simultaneously through per-agent chat tabs.

Agent Roster

Agent ID Emoji Name Discipline Conda Env Specialty
biologist 🧬 Dr. Gene Biology acaclaw-bio Genomics, sequence analysis, phylogenetics, Biopython
medscientist 🏥 Dr. Curie Medicine acaclaw-med Clinical data, survival analysis, epidemiology, DICOM
ai-researcher 🤖 Dr. Turing AI/ML acaclaw ML/DL frameworks, model training, benchmarks, arxiv
data-analyst 📊 Dr. Bayes Statistics acaclaw Pandas, R/tidyverse, visualization, statistical testing
cs-scientist 💻 Dr. Knuth Computer Science acaclaw Algorithm design, systems programming, code review

Architecture

OpenClaw Gateway (port 2090)
├── Agent: biologist    → workspace: ~/AcaClaw/agents/biologist/
│   ├── IDENTITY.md     (Dr. Gene 🧬)
│   ├── SOUL.md         (behavioral persona)
│   ├── Conda: acaclaw-bio
│   └── Session: web:main@biologist
├── Agent: medscientist → workspace: ~/AcaClaw/agents/medscientist/
│   ├── IDENTITY.md     (Dr. Curie 🏥)
│   ├── SOUL.md
│   ├── Conda: acaclaw-med
│   └── Session: web:main@medscientist
├── Agent: ai-researcher → workspace: ~/AcaClaw/agents/ai-researcher/
├── Agent: data-analyst  → workspace: ~/AcaClaw/agents/data-analyst/
└── Agent: cs-scientist  → workspace: ~/AcaClaw/agents/cs-scientist/

Each agent runs in its own session context:

  • Session key format: web:main@<agentId> — scoped per agent
  • No cross-talk: agents cannot read each other’s sessions
  • Shared data: the ~/AcaClaw/data/ directory is accessible to all agents for collaboration

How Agents Work in Parallel

  1. Per-agent chat tabs in the web UI let you send messages to different agents simultaneously
  2. Each message is routed via the session key: web:main@biologist, web:main@ai-researcher, etc.
  3. The gateway processes requests independently — one agent thinking does not block another
  4. Agents stream responses in parallel through WebSocket events scoped by runId

Parallel workflow example

You → [Dr. Gene tab]      "Analyze the RNA-seq data in data/raw/rnaseq.csv"
You → [Dr. Bayes tab]     "Run a PCA on data/processed/features.csv"
You → [Dr. Turing tab]    "Search arxiv for transformer protein models 2025-2026"

All three agents work simultaneously. Results appear in their respective tabs.

Agent Workspace Structure

Each agent gets an isolated workspace under ~/AcaClaw/agents/<id>/:

~/AcaClaw/agents/biologist/
├── IDENTITY.md          # Name, emoji, creature, vibe, theme
├── SOUL.md              # System persona and behavioral rules
├── AGENTS.md            # Workspace-specific instructions
├── memory/              # Daily memory logs
└── workspace/           # Agent's working directory
    ├── data/            # Agent-specific data
    ├── output/          # Generated results
    └── notes/           # Agent's notes

Identity and Persona

Each agent has two key files that define its character:

IDENTITY.md

Defines the visible identity — name, emoji, and visual theme:

- Name: Dr. Gene
- Emoji: 🧬
- Creature: computational biologist
- Vibe: methodical, curious, precise
- Theme: nature

SOUL.md

Defines behavioral guidelines — how the agent thinks, responds, and approaches problems:

You are a computational biologist specializing in genomics and molecular biology.
Always consider biological significance alongside statistical significance.
Prefer Biopython and scikit-bio for sequence analysis.
Use R/Bioconductor for differential expression analysis.
When presenting results, include biological context and pathway implications.

Skills per Agent

Each agent loads a skill set matching its discipline. Skills are filtered at session start.

Agent Key Skills
Dr. Gene nano-pdf, xurl, coding-agent, paper-search (biology journals)
Dr. Curie nano-pdf, xurl, coding-agent, clinical-data-tools
Dr. Turing nano-pdf, xurl, coding-agent, arxiv-search, model-benchmarks
Dr. Bayes nano-pdf, xurl, coding-agent, data-visualization
Dr. Knuth nano-pdf, xurl, coding-agent, code-review, algorithm-design

Starting Agents from the UI

  1. Navigate to Agents in the sidebar
  2. Each agent card shows status (Idle / Working), persona, and discipline
  3. Click Start on an agent card to activate it and open its chat tab
  4. The chat view shows tabs for each active agent — switch between them freely
  5. Send messages to different agents in parallel

CLI Usage

# List all agents
openclaw agents list

# Send a message to a specific agent
openclaw message --agent biologist "Analyze the FASTA sequences in data/raw/sequences.fa"

# Check agent identity
openclaw agents identity get biologist

# Start multiple agents in parallel (separate terminals)
openclaw message --agent biologist "Run sequence alignment" &
openclaw message --agent data-analyst "Generate correlation plots" &
openclaw message --agent ai-researcher "Search for RLHF papers" &
wait