# Envelope

> Envelope designs your multi-agent workflow. Describe what you need, get a complete sub-agent design — roles, skills, model, tools, and handoffs — and export without code.

Full content index (writing, blog, docs, guides, pricing, tools, roles, library, schema, MCP): https://openenvelope.org/llms-full.txt

Envelope is where multi-agent workflows get designed before they get built. Non-technical stakeholders describe what they need in plain language; Envelope structures it into a complete sub-agent design with named agents, skill definitions, model assignments, tool access, and handoff logic. The output is a structured, exportable spec — no code required to design, no ambiguity when handing off.

## Homepage

[Envelope](https://openenvelope.org/): The multi-agent designer. Users describe their AI workflow in natural language; Envelope structures it into a complete sub-agent design — roles, skills, model assignments, tool access, handoff logic, and access policies — which can be exported as a structured JSON spec. Free to start. Designs can be shared via read-only link, exported to JSON, or handed off to an engineering team for implementation.

## For agencies

[Envelope for Agencies](https://openenvelope.org/agencies): Dedicated landing page for agencies — marketing, dev shops, and consulting firms — who want to design and deliver AI agent workflows to clients. Covers the three-step flow (design → spec → hand off to client IT), use cases by agency type, and the multi-client workspace model. Positioned as the multi-agent designer for agencies building AI retainers and recurring AI workflow deliverables.

## Pricing

[Pricing](https://openenvelope.org/pricing): All plans are free during beta. Plans: Free ($0 — 5 saved designs, canvas, JSON export, share link, MCP server access), Pro ($9/mo — unlimited designs, version history, REST API access), Team ($39/mo — collaborators up to 10 seats, GitHub publish, org-wide design library), Enterprise (custom — SSO, private cloud, audit log). Founding Member lifetime deal available: $99 one-time for permanent Team access. Pricing kicks in after beta.

## Key concepts

- **Design** — the act of describing and structuring an AI agent team: naming roles, assigning tools, defining reporting lines, and setting access rules. Input is natural language; output is a structured JSON spec.
- **Spec / team definition** — the structured JSON output of a design session. Follows the open Envelope schema. Readable by engineers, storable in Git, directly implementable.
- **Agent role** — a named function in the team (e.g. "Lead Researcher", "Triage Agent") with a defined scope, tool access, model preference, and reporting relationship.
- **Tool** — a connected service or API an agent can use (HubSpot, Slack, GitHub, Stripe, and 500+ more). Tool assignments are declared in the design, credentials are provided at implementation time.
- **Reporting line** — the hierarchical relationship between agents in the team. Defines who delegates to whom and how outputs flow through the pipeline.
- **Access policy** — per-agent outbound HTTP access controls. Declared in the design; enforced at runtime by the implementing platform.
- **Human gate** — a checkpoint where a human must review or approve before the pipeline continues. Declared in the design at key decision points.
- **Version** — designs are versioned. Each published version is immutable; editors work on drafts. Existing implementations keep running until upgraded.
- **Export** — the design is exported as a JSON file following the open Envelope schema. This is the handoff artifact for IT teams.
- **Handoff** — the process of passing the exported spec to an IT or engineering team for implementation on their chosen runtime.

## Open schema

The Envelope schema is an open specification (Apache 2.0). Any platform can implement a conforming runtime. The schema `$id` is `https://schema.openenvelope.org/team/v1.json`. Designers don't need to know the schema — Envelope produces it automatically on export from the natural-language design session.

- [Schema reference](https://openenvelope.org/docs/schema): Full field-by-field reference for the `envelope-team` JSON schema — all fields, types, constraints, and examples.
- [Schema versioning policy](https://openenvelope.org/docs/schema/versioning-policy): How the schema evolves, what counts as a breaking change, version lifecycle.

## MCP & Claude Code

Envelope runs a hosted MCP server at `https://mcp.openenvelope.org/api/mcp`. Connect it to Claude.ai or ChatGPT to work with your designs from any conversation.

- [Connect to Claude.ai](https://openenvelope.org/guides/claude-ai): Add Envelope as a custom connector in Claude.ai — browse designs, export specs, and manage your team library from Claude conversations.
- [Connect to ChatGPT](https://openenvelope.org/guides/chatgpt): Add Envelope as an app in ChatGPT via OAuth. Requires ChatGPT Plus, Team, or Enterprise.
- [MCP Reference](https://openenvelope.org/mcp): Interactive reference for all MCP tools — live curl examples, auth setup, error codes.
- [Claude Code plugin](https://openenvelope.org/claude-code): Install the Envelope skill in Claude Code — teaches Claude to generate valid `.envelope.json` team definitions from plain-English descriptions.
- [MCP documentation](https://openenvelope.org/docs/mcp): Full MCP server docs — tool list, auth, install instructions for Claude.ai and ChatGPT.
- [Claude Code documentation](https://openenvelope.org/docs/claude-code): Docs for the Claude Code plugin — what it does, how to install, usage examples.

## Blog

- [From ops tool to design tool: why we pivoted](https://openenvelope.org/blog/design-tool-pivot): The story of Envelope's pivot from an AI ops platform to a design tool — why the design-first framing unlocks a bigger market, and what the product looks like now.
- [How agencies are productizing AI in 2026](https://openenvelope.org/blog/agencies-productizing-ai): Practical guide for agencies packaging AI agent workflows as named monthly retainer deliverables. Covers the leverage play, delivery problem, and how Envelope solves it with reusable design templates.
- [Run your Envelope teams from Claude and ChatGPT](https://openenvelope.org/blog/mcp-install): MCP server launch post — install steps for both clients, what each tool does, and the Claude Code skill.
- [Schema launch](https://openenvelope.org/blog/schema-launch): The launch of the open Envelope schema — the open standard for AI agent team definitions.

## Documentation

- [Schema specification](https://openenvelope.org/docs/schema): Full reference for the `envelope-team` JSON schema — all fields, types, constraints, and examples. Includes AgentDefinition, AccessPolicy, HumanGate, Schedule, and webhook event types.
- [Ecosystem overview](https://openenvelope.org/docs/ecosystem): The builder/deployer model, NPM packages, and CLI tooling for engineers implementing Envelope specs.
- [Platform integrations](https://openenvelope.org/docs/platforms): Supported runtimes for implementing an Envelope spec — Paperclip, Amazon Bedrock, Relevance AI, Envelope Managed, CrewAI, LangGraph, Vertex AI, Azure AI Foundry.
- [Versioning](https://openenvelope.org/docs/versioning): How design versioning works — creating drafts, publishing without disrupting live implementations, rolling back.
- [Access policy](https://openenvelope.org/docs/access-policy): Per-agent network access controls — allowlist/denylist rules by hostname, method, and path prefix.
- [Security & data flow](https://openenvelope.org/docs/security): What data reaches Envelope and what stays on your infrastructure.

## Guides

- [Guides overview](https://openenvelope.org/guides): Index of all setup and how-to guides.
- [Connect Claude.ai](https://openenvelope.org/guides/claude-ai): Add Envelope as a custom connector in Claude.ai — browse designs, export specs, and manage your team library from any conversation.
- [Connect ChatGPT](https://openenvelope.org/guides/chatgpt): Add Envelope as an app in ChatGPT via OAuth. Requires ChatGPT Plus, Team, or Enterprise.
- [Reading a spec](https://openenvelope.org/guides/reading-a-spec): How to read and interpret an exported Envelope team spec — fields, agent structure, and handoff logic.
- [Deploying from a spec](https://openenvelope.org/guides/deploying-from-spec): How to hand off an exported spec to an engineering team or runtime for implementation.
- [Git storage](https://openenvelope.org/guides/git-storage): How to connect a GitHub repository so designs are stored and versioned there.
- [Tool credentials](https://openenvelope.org/guides/tool-credentials): How to add credentials for connected tools (HubSpot, Slack, GitHub, etc.) so agents can access them at runtime.

## Team library

[Multi-agent team library](https://openenvelope.org/library): Browse public multi-agent team designs built on Envelope. Fork any team into your workspace and adapt it to your stack — free to start.

## Writing

Envelope publishes practical guides and essays on multi-agent design, AI workflow automation, and the case for open standards.

- [The anatomy of a multi-agent](https://openenvelope.org/writing/anatomy-multi-agent): What a multi-agent actually is — sub-agents, model routing, ordered vs. concurrent pipelines, human gates, and the design step that makes it all work. Foundational reference.
- [The stitching problem](https://openenvelope.org/writing/stitching-problem): Why most multi-agent systems fail at the connections between agents, not inside them — and why the fix is a design decision, not a better model or smarter prompt.
- [What is AI workflow automation?](https://openenvelope.org/writing/ai-workflow-automation): How AI workflow automation differs from traditional automation (Make, Zapier, n8n), why AI agents change the model, and what good multi-agent workflow design looks like.
- [How to design an AI workflow before you build it](https://openenvelope.org/writing/design-ai-workflow): Step-by-step process for designing a multi-agent workflow from objective to spec — agents, roles, tools, handoffs, pipeline structure, gates, and model assignments — before writing any code.
- [Model routing in multi-agent workflows](https://openenvelope.org/writing/model-routing): Why different agents in the same system should run on different AI models, how to match each role to the right model, and what model routing looks like in a worked example.
- [AI agents need managers too](https://openenvelope.org/writing/ai-agents-need-managers): Why governance is the infrastructure that makes AI agent delegation safe — defined scope, role-based access, audit trail, human gates.
- [The model path won't solve coordination](https://openenvelope.org/writing/model-path-governance): Better models won't solve the coordination problem. The execution layer is the governance layer.
- [Why open standards matter for AI infrastructure](https://openenvelope.org/writing/open-standards-ai-infrastructure): The case for an open definition format for AI agent teams — portability, validators, registries.
- [The platform trap in AI](https://openenvelope.org/writing/platform-trap-in-ai): Choosing an orchestration platform is a format choice. Closed formats create lock-in denser than normal infra.
- [Credentials, trust and the AI operator problem](https://openenvelope.org/writing/credentials-trust-operator-problem): The three-party credential structure, access policy as enforcement, human gates for irreversible actions.
- [Human in the loop and the autonomous agent problem](https://openenvelope.org/writing/hitl-autonomous-agent-problem): Why keeping humans in the loop matters even as agents grow more capable.
- [What makes a good agent role definition](https://openenvelope.org/writing/agent-role-definition): Roles vs prompts, agent scope and hierarchy, capabilities as constraints.
- [The composable AI thesis](https://openenvelope.org/writing/composable-ai-thesis): Why composable, narrow AI teams outperform monolithic agents.
- [Narrow by design](https://openenvelope.org/writing/narrow-by-design): The argument for building AI teams from narrow, specialised agents.

## Community

[Envelope Community](https://openenvelope.org/community): The Envelope Discord — where designers, PMs, and engineers discuss AI team design, share what they've built, and get help. Channels include #schema-discussion, #design-help, #showcase, and #platform-integrations.

## Templates by tool

Envelope has dedicated pages for every supported integration, listing community designs that use that tool.

- [HubSpot AI teams](https://openenvelope.org/tools/hubspot)
- [Slack AI teams](https://openenvelope.org/tools/slack)
- [Zendesk AI teams](https://openenvelope.org/tools/zendesk)
- [Salesforce AI teams](https://openenvelope.org/tools/salesforce)
- [GitHub AI teams](https://openenvelope.org/tools/github)
- [Linear AI teams](https://openenvelope.org/tools/linear)
- [Stripe AI teams](https://openenvelope.org/tools/stripe)
- [Notion AI teams](https://openenvelope.org/tools/notion)
- [Jira AI teams](https://openenvelope.org/tools/jira)
- [Gmail AI teams](https://openenvelope.org/tools/gmail)
- [Intercom AI teams](https://openenvelope.org/tools/intercom)
- [Freshdesk AI teams](https://openenvelope.org/tools/freshdesk)

## Role landing pages

Envelope has pages for common agent roles in AI team designs.

- [Sales roles](https://openenvelope.org/roles/sales)
- [Support roles](https://openenvelope.org/roles/support)
- [Engineering roles](https://openenvelope.org/roles/engineering)
- [Marketing roles](https://openenvelope.org/roles/marketing)
- [Operations roles](https://openenvelope.org/roles/operations)
- [Finance roles](https://openenvelope.org/roles/finance)
- [HR roles](https://openenvelope.org/roles/hr)