MESS-Agents

This repository is a downstream mirror. Source of truth lives in the messai-ai monorepo; this mirror is updated on each release. Issues and Discussions are welcome here. PRs against this mirror will be redirected — see CONTRIBUTING.md.

History was reset as part of the 2026 monorepo consolidation. Versions tagged before that (e.g. v0.2.0) remain accessible as historical refs.

Multi-agent research orchestration framework for MES research

Overview

MESS-Agents provides a multi-agent framework for orchestrating research workflows:

• Base Agent Framework - Abstract agent class for custom agents
• Agent Orchestrator - Multi-agent coordination patterns
• Literature Analyzer - Paper analysis patterns
• Insights Generator - Finding synthesis
• Knowledge Graph Builder - Graph construction
• PDF Processor - Document processing

Installation

Not yet published to npm. This package is source-available here while its public API stabilises. Use it by cloning the mirror:

git clone https://github.com/Messai-io/MESS-Agents.git
cd MESS-Agents && pnpm install && pnpm build

Track the packaging issue for the npm release.

Features

Base Agent Class

import { BaseAgent, AgentConfig } from '@messai-io/mess-agents';

class CustomResearchAgent extends BaseAgent {
  constructor(config: AgentConfig) {
    super({
      name: 'CustomResearcher',
      capabilities: ['literature_search', 'data_analysis'],
      ...config,
    });
  }

  async execute(task: Task): Promise<Result> {
    // Custom implementation
    const papers = await this.searchLiterature(task.query);
    const analysis = await this.analyze(papers);
    return { findings: analysis };
  }
}

Agent Orchestrator

import { AgentOrchestrator } from '@messai-io/mess-agents';

const orchestrator = new AgentOrchestrator({
  maxConcurrency: 5,
  timeout: 60000,
  costOptimization: true,
});

// Register agents
orchestrator.register(new LiteratureAgent());
orchestrator.register(new AnalysisAgent());
orchestrator.register(new SynthesisAgent());

// Execute workflow
const results = await orchestrator.execute({
  task: 'Analyze biofilm conductivity in MFCs',
  workflow: 'literature_review',
});

Literature Analyzer

import { LiteratureAnalyzer } from '@messai-io/mess-agents';

const analyzer = new LiteratureAnalyzer({
  maxPapers: 100,
  dateRange: { start: '2020-01-01', end: '2025-01-01' },
});

const analysis = await analyzer.analyze({
  query: 'microbial fuel cell power density',
  focusAreas: ['materials', 'performance', 'scale-up'],
});

console.log(analysis.keyFindings);
console.log(analysis.researchGaps);
console.log(analysis.trends);

Insights Generator

import { InsightsGenerator } from '@messai-io/mess-agents';

const generator = new InsightsGenerator();

// Synthesize findings from multiple sources
const insights = await generator.synthesize({
  sources: [literatureAnalysis, experimentalData, modelPredictions],
  focusQuestion: 'What limits MFC power output?',
});

console.log(insights.mainConclusions);
console.log(insights.confidenceLevel);
console.log(insights.supportingEvidence);

Knowledge Graph Builder

import { KnowledgeGraphBuilder } from '@messai-io/mess-agents';

const builder = new KnowledgeGraphBuilder();

// Build knowledge graph from papers
const graph = await builder.build({
  papers: analyzedPapers,
  entityTypes: ['material', 'organism', 'parameter', 'application'],
  relationTypes: ['uses', 'produces', 'inhibits', 'enhances'],
});

// Query the graph
const related = graph.query({
  entity: 'Geobacter',
  relation: 'produces',
  depth: 2,
});

PDF Processor

import { PDFProcessor } from '@messai-io/mess-agents';

const processor = new PDFProcessor({
  extractFigures: true,
  extractTables: true,
  extractEquations: true,
});

const document = await processor.process('paper.pdf');

console.log(document.text);
console.log(document.figures);
console.log(document.tables);
console.log(document.references);

Workflow Templates

Pre-built workflow templates for common research tasks:

import { Workflows } from '@messai-io/mess-agents';

// Literature review workflow
const review = await Workflows.literatureReview({
  topic: 'MFC cathode materials',
  depth: 'comprehensive',
});

// Gap analysis workflow
const gaps = await Workflows.gapAnalysis({
  field: 'bioelectrochemical_systems',
  timeRange: '2020-2025',
});

// Hypothesis generation workflow
const hypotheses = await Workflows.generateHypotheses({
  observations: experimentalResults,
  priorKnowledge: literatureFindings,
});

Cost Optimization

Built-in cost optimization for LLM usage:

const orchestrator = new AgentOrchestrator({
  costOptimization: {
    enabled: true,
    maxCostPerTask: 1.0, // USD
    preferLocalModels: true,
    cacheResults: true,
  },
});

API Reference

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Links

• MESSAI Platform
• Documentation
• Examples