MYRIAD

A 10,000-node biotech agent skill graph

100 domains / 1,000 workstreams / 10,000 atomic task nodes

Built by Srihari Muralikrishnan with ChatGPT for Kalaris Labs

Try it in 10 seconds

npx kalaris-myriad stats

MYRIAD 1.2.0
100 domains / 1,000 workstreams / 10,000 tasks / 10 batches
Maturity: taxonomy-defined; full seed skills: 3

Search the graph:

npx kalaris-myriad search "variant-aware off-target" --limit 5

Inspect a node:

npx kalaris-myriad show MYR-D051-T041

Export a domain:

npx kalaris-myriad export \
  --domain 21 \
  --format md \
  --output protein-structure-prediction.md

What MYRIAD is

MYRIAD is a release-engineered, machine-readable map of work across biotechnology and pharmaceutical development.

It turns broad fields such as genome assembly, protein engineering, drug discovery, bioprocessing, clinical development, and regulatory science into bounded, searchable agent tasks.

Each task node provides a stable routing surface for:

• agent discovery and delegation;
• retrieval-augmented planning;
• workflow decomposition;
• evaluation and benchmarking;
• conversion into complete production SKILL.md implementations.

What MYRIAD is not

MYRIAD is not a collection of 10,000 experimentally validated wet-lab protocols.

The graph currently contains:

Operational use still requires current primary sources, validated local procedures, appropriate authorization, and qualified human review.

The graph at a glance

100 domains
  x 10 workstreams per domain
  x 10 atomic tasks per workstream
  = 10,000 task nodes

Coverage

Molecular and omics

• Genome assembly and pangenomics
• Germline, somatic, and structural variation
• Population and rare-disease genomics
• Bulk, single-cell, and spatial transcriptomics
• Epigenomics and multi-omics
• Proteomics, PTM analysis, and metabolomics

Structure and discovery

• Protein structure prediction
• Cryo-EM, crystallography, and NMR
• Molecular dynamics and QM/MM
• Protein and antibody engineering
• Target identification and validation
• Virtual screening, docking, QSAR, and free energy

Therapeutics and development

• Medicinal chemistry, ADME, DMPK, and PK-PD
• Predictive toxicology and biomarkers
• CRISPR, base editing, and prime editing
• Gene therapy, RNA therapeutics, and delivery
• Antibodies, ADCs, degraders, vaccines, and cell therapy

Manufacturing and translation

• Cell-line and upstream process development
• Bioreactor control, PAT, QbD, and purification
• Formulation, stability, and comparability
• Microbial, synthetic, agricultural, and environmental biotech
• Clinical trials, RWE, pharmacovigilance, GxP, FAIR data, and AI governance

Use MYRIAD

Explore in a browser

Live explorer: https://myriad-explorer.vercel.app

The dependency-free explorer reads the canonical graph directly and provides search, domain/workstream filtering, shareable task URLs, JSON export, task/CLI copy actions, and links to canonical GitHub nodes.

npm run explorer

Then open http://localhost:4173/explorer/. A local server is required so the browser can load the canonical JSONL data.

Run without installing

npx kalaris-myriad <command>

Install globally

npm install --global kalaris-myriad
kalaris-myriad stats

Clone for local development

git clone https://github.com/KalarisLabs/myriad.git
cd myriad
npm ci
npm test
npm run validate

CLI reference

Search only the bioreactor domain:

npx kalaris-myriad search "residence-time distribution" --domain 78

Return machine-readable output:

npx kalaris-myriad show MYR-D078-T001 --json

Export one domain to JSONL:

npx kalaris-myriad export \
  --domain 51 \
  --format jsonl \
  --output crispr-guide-design.jsonl

Validate a globally installed release:

kalaris-myriad validate --json

One graph, three representations

MYRIAD preserves three layers deliberately:

Earlier nodes were not padded with invented scientific detail merely to make every batch look equally verbose. Source meaning is preserved, while the canonical layer normalizes operational boundaries.

Repository map

.
|-- bin/                 Dependency-free command-line interface
|-- data/
|   |-- raw/             Preserved source records
|   `-- canonical/       Homogenized 10,000-node graph
|-- dist/sanitized/      Public-facing distribution
|-- docs/                Architecture, QA, safety, and release documentation
|-- lib/                 CLI and data-access implementation
|-- myriad/
|   |-- batches/         Ten standardized batches
|   `-- domains/         100 domains and 10,000 node Markdown files
|-- references/          Controlled reference documentation
|-- schemas/             JSON Schema contracts
|-- scripts/             Release and scientific-structure validators
|-- skills/              Complete seed SKILL.md implementations
|-- tests/               Deterministic Node.js tests
|-- CITATION.cff
|-- CREDITS.md
|-- LICENSE.md
|-- NOTICE.md
|-- SECURITY.md
`-- package.json

Seed skills

Three nodes have been expanded into complete reference implementations:

1. Diploid PacBio HiFi assembly optimization
2. Variant-aware CRISPR off-target scoring
3. HTS normalization, plate QC, and hit calling

These demonstrate the target shape for Phase 3: strict frontmatter, input schemas, procedural stages, decision gates, provenance, failure handling, outputs, and human-review boundaries.

Design principles

Precise routing

Task names communicate when a node should activate and what decision artifact it produces.

Progressive disclosure

Agents can begin with the global index, narrow to a domain or workstream, and load a full node only when needed.

Evidence before confidence

A score, prediction, or model output is never treated as experimental, clinical, manufacturing, or regulatory proof.

Fail closed

Nodes support explicit pass, warning, fail, and no-call outcomes rather than forcing unsupported conclusions.

Human authority remains explicit

The graph does not authorize autonomous wet-lab execution, clinical decisions, GMP disposition, regulatory determinations, or safety overrides.

Validation

The release validator checks:

• exactly 100 domains and 10,000 task nodes;
• exactly 100 tasks per domain;
• globally unique IDs, routing names, titles, and objectives;
• valid JSONL and node Markdown frontmatter;
• required execution boundaries and completion evidence;
• placeholder and near-duplicate detection;
• schema compliance;
• deterministic CLI behavior;
• npm package contents.

Run everything locally:

npm ci
npm test
npm run validate
npm run pack:check

See docs/QUALITY_ASSURANCE_REPORT.md for the readable release summary.

Safety boundary

MYRIAD is computational-advisory infrastructure.

It must not autonomously:

• execute wet-lab procedures;
• operate or alter bioreactor controls;
• prescribe treatment or determine patient care;
• release clinical or diagnostic results;
• disposition GMP lots;
• issue final regulatory-compliance determinations;
• bypass institutional biosafety, biosecurity, ethics, privacy, or authorization controls.

A node is not a substitute for current primary literature, validated protocols, local SOPs, approved lab systems, or qualified professional judgment.

Read docs/SAFETY_AND_LIMITATIONS.md.

Build with MYRIAD

Good contributions include:

• implementing a taxonomy node as a complete SKILL.md;
• improving source mappings and version resolution;
• adding reproducible evaluation fixtures;
• correcting scientific scope or terminology;
• improving routing, schemas, validators, or documentation.

Please read CONTRIBUTING.md before opening a pull request.

Do not submit credentials, proprietary datasets, controlled information, unsupported clinical recommendations, or operational procedures you are not authorized to disclose.

Citation

Citation metadata is available in CITATION.cff.

Suggested acknowledgement:

Srihari Muralikrishnan and Kalaris Labs. MYRIAD: A 10,000-Node Biotech Agent Skill Graph. Version 1.0.2, 2026. AI systems architecture and release engineering supported by ChatGPT, OpenAI.

Credits

• Srihari Muralikrishnan - creator, project lead, systems architect, and co-founder of Kalaris Labs.
• ChatGPT - GPT-5.6 Thinking by OpenAI - AI systems architecture, synthesis, normalization, validation tooling, documentation, and release engineering.
• Kalaris Labs - project home, organizational direction, and release stewardship.

See AUTHORS.md and CREDITS.md.

License

MYRIAD uses a split, attribution-preserving license:

• Software and executable tooling: Apache License 2.0
• Task graph, datasets, and documentation: CC BY 4.0

Public use, modification, redistribution, and commercial use are allowed under the applicable license. Preserve the required notices and credit:

MYRIAD by Srihari Muralikrishnan and Kalaris Labs.
https://github.com/KalarisLabs/myriad

See LICENSE.md for the file-by-file scope and attribution requirements.

Search the graph. Inspect the node. Build the skill.

npx kalaris-myriad search "your biotech problem"

Kalaris Labs / Srihari Muralikrishnan / ChatGPT