SKILL.md · Claude Code, Codex & 30 more tools

The only
framework
built for production

Compliance, token governance, durable execution, and audit trails built in — not bolted on. Ship production-safe AI agents in 7 lines of Python.

Apache 2.0 — forever5,711 tests passingPython 3.11+HIPAA compliant

workflow.py

# 7 lines. Production-safe by default.

from meshflow import Workflow, Agent, CostCap

wf = Workflow(cost_cap=CostCap(usd=5.00))

wf.add(Agent('researcher'), Agent('analyst'), Agent('writer'))

result = wf.run('Write a competitive analysis')

# Compliant. Durable. Audited. Cost-capped.

Cloud model tasks

~10%

rest handled by local tiers

Governance checks

per step, every run

OSSApache 2.0 — free forever, no platform tax, no license changes.The full framework, all guardrails, 21 agents, SKILL.md, and sandbox mode are free under a standard OSS license. No seat limits. No workflow limits. No expiry.

All 9 guardrail types21 specialist agentsSHA-256 audit chainMCP Server auto-genSandbox modeSKILL.md (32 tools)

$10.8BAgent market in 2026

68ptAdoption vs production gap

~10%Tasks needing cloud models

32AI tools via SKILL.md

5 ★Unique advantages

21Specialist agents built-in

$0Platform tax, forever

7Lines of code to production

$10.8BAgent market in 2026

68ptAdoption vs production gap

~10%Tasks needing cloud models

32AI tools via SKILL.md

5 ★Unique advantages

21Specialist agents built-in

$0Platform tax, forever

7Lines of code to production

The production problem

79% of enterprises adopt AI agents.
Only 11% ship them to production.

79% — adopt AI agents11% — run in production

Source: McKinsey State of AI 2024 · adoption vs. production deployment gap across enterprise respondents.

Every other framework

Compliance is your engineering problem

HIPAA safeguards? Write them yourself

SHA-256 audit chain? Build from scratch

Token costs spiral undetected at scale

No cost caps — you find out at billing

MeshFlow

ComplianceGuard ships day one, zero config

HIPAA + SOC 2 enforced at the workflow layer

SHA-256 audit chain on every agent step

ModelRouter routes ~90% of tasks to local models

Hard spending limits before agents run

Built-in capabilities

Everything production
needs. Nothing to configure.

The default path is the safe path. Safety features are built in, not bolted on.

ComplianceGuard

HIPAA, SOC 2, GDPR controls enforced at the workflow layer. PII detection, SensitiveDataDetector, PolicyGuard, and a SHA-256 audit chain ship by default.

wf = Workflow() # ComplianceGuard auto-applied

★ Unique to MeshFlow

70%

Token Governance

CostCapGuardrail, ModelRouter, ContextCompactor, RAG budgets.

★ Unique

DurableWorkflowExecutor

Workflows survive crashes and restarts. Resume exactly where they stopped — no re-execution, no data loss.

★ Unique

MCP Server Auto-Gen

Any MeshFlow workflow becomes an API endpoint automatically. Other agents call your workflows as services.

★ Unique

Multi-Agent Orchestration

GroupChat, HITL interrupt points, 21 pre-built specialist agents. Production-grade from day one.

Built-in

Sandbox Mode

mode='sandbox' — full workflows, mocked LLMs, zero tokens spent.

Developer-first

Cross-Session Memory

Long-term memory with Weaviate / Pinecone. Agents remember across conversations.

Built-in

OTEL + Cost Analytics

Full OpenTelemetry. TokQ cost analytics per workflow. Prometheus-ready. Zero platform tax.

Built-in

Competitive landscape

The only framework that ships
production safety as a default.

★5

Capability	LangGraph	CrewAI	AutoGen	Dify	Flowise	Haystack	MeshFlow
Multi-agent orchestration	✅	✅	✅	⚡	⚡	✗	✅
Compliance (HIPAA/SOC2)	✗	✗	✗	⚡	✗	✗	✅ ★
Built-in guardrails	✗	✗	✗	✗	✗	✗	✅ ★
Token cost governance	✗	✗	✗	✗	✗	✗	✅ ★
Durable execution / resume	✅	⚡	✗	✗	✗	✗	✅ ★
SHA-256 audit chain	✗	✗	✗	✗	✗	✗	✅ ★
MCP Server auto-gen	✗	✗	✗	✗	✗	✗	✅ ★
Cross-session memory	⚡	⚡	✗	✗	✗	✗	✅
HITL (human-in-the-loop)	✅	✅	⚡	✗	✅	✗	✅

★ = MeshFlow unique advantage — not available in any competitor framework

Universal Distribution

One SKILL.md.
32 AI coding tools.
Auto-invoked.

Drop meshflow/SKILL.md into your project and MeshFlow auto-invokes across every major AI coding tool. The fastest cross-vendor standardization in AI tooling history.

meshflow/SKILL.md

---

name: meshflow

description: >

Invoke when user wants multi-agent workflows,

durable execution, compliance, or cost caps.

slash_command: /meshflow

context: fork

---

# Production-safe defaults applied

# Zero configuration required

Claude Code

Codex CLI

Cursor

Gemini CLI

GitHub Copilot

JetBrains Junie

AWS Kiro

Windsurf

Continue.dev

Aider

Cline

Bolt.new

Replit

Devin

Zed AI

Supermaven

Tabnine

Cody

Phind

Perplexity

Poe

You.com

Pear AI

Pythagora

GPT Engineer

Sweep

SWE-agent

OpenDevin

AgentCoder

AutoCodeRover

Factory.ai

Claude Code

Codex CLI

Cursor

Gemini CLI

GitHub Copilot

JetBrains Junie

AWS Kiro

Windsurf

Continue.dev

Aider

Cline

Bolt.new

Replit

Devin

Zed AI

Supermaven

Tabnine

Cody

Phind

Perplexity

Poe

You.com

Pear AI

Pythagora

GPT Engineer

Sweep

SWE-agent

OpenDevin

AgentCoder

AutoCodeRover

Factory.ai

Token Optimization ROI

The savings pay for implementation
in month one.

Five token cost levers, all enforced at the kernel level. No competitor has any of them.

Scenario 01

10M tokens/day production agent

$210K+

Annual savings vs. unoptimized LLM spend

$300K/yr~$75K/yr with MeshFlow

Scenario 02 — Most common

100-agent parallel crew (20-step workflow)

60–85%

Context dedup + ModelRouter brings crew cost near single-agent baseline

4–15× baseline~1× with MeshFlow

Scenario 03

HIPAA long-context audit logging

70–90%

cache_control on system prompts: 10% of normal price on cached tokens

Full re-send/turnCached at 10% cost

Built for teams across healthcare, finance, and legal

Healthcare ops

Financial services

Legal teams

AI platforms

Internal tools

Support automation

Auditable by design

Healthcare & Life Sciences

Problem: HIPAA requires evidence that PHI never reached a model call unguarded, and that every agent action is auditable.

How MeshFlow handles it: PIIBlocker runs before every LLM call. SHA-256 hash-chained StepRecords are written for every agent step. compliance_profile("hipaa") activates 6-year retention policy and access-log generation in one line.

Hard cap, not a soft limit

Financial Services & FinOps

Problem: AI spend becomes opaque across teams. Cost caps set in code get bypassed in practice.

How MeshFlow handles it: CostCap(usd=N) is enforced in StepRuntime — the kernel that every agent step passes through. It cannot be bypassed. ModelRouter routes ~90% of tasks to local models; only ~10% reach paid cloud APIs.

Hand to an auditor directly

Legal & Compliance Teams

Problem: Procurement and security need reproducible proof that agents behaved within policy — not a trace they have to interpret.

How MeshFlow handles it: ReplayLedger stores every run with prev_hash/entry_hash for tamper-evidence. SOC2Checker maps 18 controls to AICPA Trust Services Criteria. meshflow-forensic generates auditor-ready HTML reports.

Launch notes for production teams

LangGraph migration

Bring graph-style orchestration into a framework that treats audit trails, compliance profiles, and cost controls as first-class workflow behavior.

Guide

Parallel crews

Run multi-agent workflows with shared context, per-step cost visibility, and routing decisions that can be inspected after execution.

Guide

Sandbox mode

Validate workflow shape, handoffs, and expected output quality before connecting paid model calls or production data.

Guide

Flowise replacement

Move visual workflow experiments into reviewable Python when security, deployment, and maintainability become the priority.

Guide

Compliance review

Give legal and security teams structured execution evidence instead of asking them to infer safety from raw traces.

Guide

SKILL.md workflow

Let coding agents scaffold MeshFlow patterns automatically when developers ask for agents, workflows, or orchestration.

Guide

Get started today

The golden standard
starts here.

Production-safe agents in 7 lines. Free forever, open source, no platform tax.

$pip install meshflow

Apache 2.0 — forever, no license changes

HIPAA-compliant architecture

5,711 tests passing · Production/Stable

Works with Claude Code, Codex & Cursor

79% of enterprises adopt AI agents.
Only 11% ship them to production.

79% — adopt AI agents11% — run in production

Source: McKinsey State of AI 2024 · adoption vs. production deployment gap across enterprise respondents.

Every other framework

Compliance is your engineering problem

HIPAA safeguards? Write them yourself

SHA-256 audit chain? Build from scratch

Token costs spiral undetected at scale

No cost caps — you find out at billing

MeshFlow

ComplianceGuard ships day one, zero config

HIPAA + SOC 2 enforced at the workflow layer

SHA-256 audit chain on every agent step

ModelRouter routes ~90% of tasks to local models

Hard spending limits before agents run

Capability

LangGraph

CrewAI

AutoGen

Dify

Flowise

Haystack

MeshFlow

Multi-agent orchestration

✅

⚡

✗

✅

Compliance (HIPAA/SOC2)

✗

⚡

✗

✅ ★

Built-in guardrails

✗

✅ ★

Token cost governance

✗

✅ ★

Durable execution / resume

✅

⚡

✗

✅ ★

SHA-256 audit chain

✗

✅ ★

MCP Server auto-gen

✗

✅ ★

Cross-session memory

⚡

✗

✅

HITL (human-in-the-loop)

✅

⚡

✗

✅

✗

✅

The savings pay for implementation
in month one.

Five token cost levers, all enforced at the kernel level. No competitor has any of them.

Scenario 01

10M tokens/day production agent

$210K+

Annual savings vs. unoptimized LLM spend

$300K/yr~$75K/yr with MeshFlow

Scenario 02 — Most common

100-agent parallel crew (20-step workflow)

60–85%

Context dedup + ModelRouter brings crew cost near single-agent baseline

4–15× baseline~1× with MeshFlow

Scenario 03

HIPAA long-context audit logging

70–90%

cache_control on system prompts: 10% of normal price on cached tokens

Full re-send/turnCached at 10% cost

The onlyframeworkbuilt for production

79% of enterprises adopt AI agents.Only 11% ship them to production.

The only framework that shipsproduction safety as a default.

One SKILL.md.32 AI coding tools.Auto-invoked.

The savings pay for implementationin month one.

The golden standardstarts here.

The onlyframeworkbuilt for production

79% of enterprises adopt AI agents.Only 11% ship them to production.

The only framework that shipsproduction safety as a default.

One SKILL.md.32 AI coding tools.Auto-invoked.

The savings pay for implementationin month one.

The golden standardstarts here.

The only
framework
built for production

79% of enterprises adopt AI agents.
Only 11% ship them to production.

The only framework that ships
production safety as a default.

One SKILL.md.
32 AI coding tools.
Auto-invoked.

The savings pay for implementation
in month one.

The golden standard
starts here.

The only
framework
built for production

79% of enterprises adopt AI agents.
Only 11% ship them to production.

The only framework that ships
production safety as a default.

One SKILL.md.
32 AI coding tools.
Auto-invoked.

The savings pay for implementation
in month one.

The golden standard
starts here.