Choosing a platform

Envelope deploys teams to any supported orchestration platform. This guide covers which platforms are available, what credentials each one requires, and how to pick the right target for your use case.

Looking for concept mapping, individual credential reference, or integration status? Platform integration docs has a page for each platform with full technical detail.

What is a platform?

An orchestration platform is where your AI agents actually live and run after deployment. Envelope holds the team definition — the roles, hierarchy, prompts, and pricing. When you deploy, Envelope reads that definition and translates it into whatever format the target platform expects, then provisions all agents automatically.

The same team definition works across all platforms. The creator doesn't need to know or care which platform the deployer uses — they publish once, and deployers choose their runtime.

Currently supported

Envelope Managed

Envelope cloud · live

Envelope Managed is Envelope's own hosted runtime — no external account, no credentials, no infrastructure to configure. You deploy directly onto Envelope's cloud and agents run there immediately. The simplest way to get a team running.

No credentials required — Envelope Managed installs do not require any credential fields. Your Envelope API key is the only authentication needed — Envelope handles the rest.

Observability — full token data. Because the runtime is native to Envelope, every run is captured automatically. Both per-run and per-1K-token billing are fully supported.

---

Paperclip

Self-hosted · live

Paperclip is a self-hosted orchestration platform — you run the server, you own the data. Envelope provisions agents and org structure via its REST API. Best for teams that already operate a Paperclip instance or want on-premise agent infrastructure.

Required credentials:

Auto-company provisioning — pass a company name instead of a company ID and Envelope will create a new org in Paperclip and deploy into it in a single step.

Observability — fully supported. Envelope installs the observability plugin automatically at deploy time. It fires on every completed run and reports token counts back to Envelope — both per-run and per-1K-token billing work.

---

Relevance AI

Managed cloud · live

Relevance AI is a managed cloud platform — no infrastructure to run. Agents are deployed directly into your Relevance AI project workspace with the team's hierarchy wired up as sub-agent tool references. Best for teams that want immediate cloud deployment with no ops overhead.

Required credentials:

Where to find your credentials — log in to app.relevanceai.com, go to Settings → API Keys. Your Project ID, API Key, and Region code are all on that page. Generate an API key with Admin role.

Using Claude or Gemini on Relevance AI — Relevance AI only includes OpenAI access by default. To use Anthropic or Google models, the deployer must add their own API key in the Relevance AI workspace: Settings → API Keys → Add Provider Key. Agents will install successfully regardless — this only affects invocation.

Observability — run counts only (no token data). Relevance AI does not expose token usage via its API. Only per-run billing is supported — per-1K-token pricing will not generate billing events for Relevance AI deployments.

---

Amazon Bedrock

Managed cloud (AWS) · live

Amazon Bedrock Agents is AWS's managed orchestration platform. Envelope deploys teams using Bedrock's supervisor/collaborator model — one root agent supervises all others. Best for teams already running workloads on AWS who want agents inside their own AWS account.

Hub-and-spoke only — one level deep. Bedrock's multi-agent model supports exactly one level of delegation. Envelope will hard-fail at deploy time if your team has more than two tiers of reporting relationships, with a clear error message before any AWS calls are made.

Required credentials:

IAM setup — two objects required:

1. Agent Runtime Role — create an IAM role trusted by bedrock.amazonaws.com with bedrock:InvokeModel permission. This role is assumed by each agent at runtime. Create it once and reuse across all deploys.

2. Caller credentials — an IAM user (or role) with permissions to create, prepare, and invoke Bedrock agents, associate collaborators, and pass the runtime role. Generate an Access Key under IAM → Users → Security credentials.

Production recommendation — IAM role assumption. For production deployments, prefer creating a dedicated IAM role in your account and granting Envelope permission to assume it, rather than providing long-lived static credentials. Contact your AWS administrator to set up a cross-account trust relationship with Envelope's AWS account ID.

Model compatibility & enablement — Bedrock only hosts models from its own catalogue — OpenAI and Google models are not available. Supported families include Anthropic Claude, Amazon Titan and Nova, Meta Llama, Mistral, Cohere, and AI21. Before you deploy, the platform compatibility strip will show amber if any agent in your team uses an unsupported model.

Required before deploying: models must be explicitly enabled in your AWS account. Open the Bedrock console → Model access → Request access for each model your team uses. Deploying without this will fail with an access denied error from AWS.

Observability — full token data. Invoke deployed teams through Envelope's proxy endpoint — no Bedrock SDK required. Envelope relays the call, captures token counts from the response, and records the run automatically. Both per-run and per-1K-token billing are fully supported.

VPC endpoint note — if your agents need to call Bedrock through a private VPC endpoint (common in air-gapped enterprise environments), invoke Bedrock directly from your own code within the VPC and self-report runs to Envelope.

Teardown & cleanup — when you no longer need a Bedrock install, use the teardown endpoint. Envelope will delete all provisioned agents and their aliases from your AWS account, remove the stored credentials, and mark the install as cancelled.

Important: AWS charges for active Bedrock agents. Always tear down installs you no longer need — do not simply stop calling them.

---

Coming soon

Listed in integration priority order — conceptual fit first, then enterprise tier. Each will appear in the platform picker once integration is complete.

CrewAI Enterprise (Managed cloud) — closest conceptual match — crew hierarchy maps directly to Envelope team structure. Research complete.

LangGraph Cloud (Managed cloud) — unlimited hierarchy depth via subgraph nesting. Supervisor-routing pattern aligns with Envelope teams. Research complete.

Vertex AI Agent Builder (Managed cloud) — unlimited hierarchy depth via ADK sub-agent tree. Research complete.

Azure AI Agent Service (Managed cloud) — OpenAI Assistants-compatible REST API. Hierarchy approximated via Connected Agents. Research complete.

How to choose

Pick Envelope Managed if:

• You want the fastest path from team to running agents
• You have no existing orchestration platform
• You're prototyping or evaluating before committing to infrastructure

Pick Paperclip if:

• You need on-premise or self-hosted agents
• You already operate a Paperclip instance
• You want multi-company org support within a single deployment

Pick Relevance AI if:

• You don't want to run any infrastructure yourself
• You want a managed cloud workspace out of the box
• You're deploying for a single team or project

Pick Amazon Bedrock if:

• Your workloads already run on AWS
• You need agents inside your own AWS account
• Your team hierarchy is flat (one supervisor, direct reports only)

Discover platforms via API

Call the platforms API endpoint to get the current list of supported platforms and their required credential fields. This is useful for building dynamic deployment interfaces that automatically adapt as new platforms are added. See the full API reference.