Sandbox
@speakeasy-api/gram

AI control plane for agents, MCPs, and skills

Gram is a full control plane for company-wide AI usage, not a small add-on. It connects agents to MCPs, applies permissions and policy checks, records usage and costs, and distributes skills and assistants through the organization.

266 stars31 forksGoUpdated 6d ago
Who it's for

Builders who need a central system to govern how their team uses agents, MCPs, and skills.

What it delivers

You can manage AI access, audit tool use, and roll out shared agent resources from one place.

What it does

Observe AI usage

Tracks token and cost use, with searchable logs for sessions, tool calls, permission changes, and access events.

Secure prompts and actions

Inspects prompts, responses, and agent actions in real time to block, redact, or log sensitive data.

Connect to MCPs

Provides a control layer for connecting agents to SaaS tools, APIs, and internal systems with policy enforcement.

Distribute shared resources

Rolls out MCPs, skills, plugins, and assistants by role with RBAC and Oauth2.1.

Local development stack

Uses `./zero` to install dependencies, run migrations, and start the dashboard and services.

How to get it

  1. 1Seed your local organization with a realistic working environment — a deployed API,…
    mise seed
  2. 2To work on several branches at once, each git worktree can run its own full stack on its…
    wt switch --create my-feature

README

Gram by Speakeasy

Speakeasy AI Control Plane

Securely scale AI usage across your organization. Built for humans and agents.
Learn more »

Documentation · Running locally · Tech Stack · Contributing · Login · Roadmap

Introduction

Gram is the open source stack behind Speakeasy's AI control plane. Secure and centrally manage MCPs, Skills, and Assistants your whole company to access, with fine-grained permissions, threat detection, and full observability of token use and costs. Every tool call, permission change, and access event logged and searchable. SOC 2 Type II and ISO 27001 certified.

To get started on the hosted platform you can Sign up, or check out the Quickstart guide.

Supports popular AI providers

Claude      Claude Code      ChatGPT      Codex      Gemini      Cursor      GitHub Copilot

Observe

Track AI usage across teams and measure impact with either tokens or cost. Deep dive expensive sessions, create budgets and measure tool effectiveness. Built on a foundation of Opentelemetry. Exportable and interactive via platform MCP and a built in assistant.

Secure

Every prompt, response, and agent action is inspected and enforced in real time. Sensitive data is blocked, redacted, or logged before it leaves your environment. Create and enforce flexible policies to prevent prompt injection, log shadow use and detect PII secrets and other sensitive data types.

Connect

A single control layer for connecting your agents to MCPs for your SaaS vendors, APIs, and internal systems, with policy enforcement and granular access control built in. Private networking and tunneling deployed on demand.

Distribute

Centralise distribution of MCPs, Skills, Plugins and Assistants to your team based on enterprise roles. Team, server, and tool level permissions enforced through RBAC and Oauth2.1. Synced to your enterprise IDP (Okta, Azure AD, Google Workspace, etc.).

Support

Running locally

Run ./zero until it succeeds. This script is what you use to run the dashboard and services for local development. It installs dependencies, runs pending database migrations, and starts everything up.

The main dependencies are Mise and Docker. The ./zero script will guide you to install these if they are not found.

Seed your local organization with a realistic working environment — a deployed API, agent sessions, telemetry, risk findings, teammates, and an API key:

mise seed

It talks to the databases directly, so it can run before the services are up, and re-running it resets the seeded data.

To work on several branches at once, each git worktree can run its own full stack on its own ports:

wt switch --create my-feature

See CONTRIBUTING.md for more detail on local development, auth, worktrees, and the CLI.

Contributing

Contributions are welcome! Please open an issue or discussion for questions or suggestions before starting significant work.

See CONTRIBUTING.md for development setup and detailed contribution guidelines.

Tech Stack

Contributors

Gram contributors

Built by Speakeasy

Files in the repo

Repository payload86 top-level entries
  • .agents
  • .changeset
  • .claude
  • .codex
  • .config
  • .cursor
  • .github
  • .impeccable
  • .mise-tasks
  • .opencode
  • .playwright
  • .speakeasy
  • .vscode
  • agents
  • assets
  • ci
  • cli
  • client
  • dev-idp
  • dev-idp-dashboard
  • docs
  • e2e
  • examples
  • functions
  • glint
  • hooks
  • infra
  • local
  • mock-oidc
  • otel-collector
  • overlays
  • plog
  • pystreams
  • redirects
  • seed
  • server
  • skills
  • tools
  • ts-framework
  • tunnel
  • .gitattributes
  • .gitignore
  • .goreleaser.hooks.yaml
  • .goreleaser.yaml
  • .mcp.json
  • .node-version
  • .oxfmtrc.json
  • .playwright-mcp.config.json
  • .worktreeinclude
  • AGENTS.md
  • buf.gen.yaml
  • buf.yaml
  • CLA.md
  • CLAUDE.md
  • CODEOWNERS
  • compose.shared.yml
  • compose.yml
  • config.template.toml
  • CONTRIBUTING.md
  • cubic.yaml
  • GEMINI.md
  • go.mod
  • go.sum
  • hk.pkl
  • install-cli.ps1
  • install-cli.sh
  • LICENSE
  • mise.lock
  • mise.toml
  • mise.viteprod.toml
  • mprocs.yaml
  • ONBOARDING.md
  • opencode.json
  • package.json
  • pin-hooks-release
  • pitchfork.toml
  • pnpm-lock.yaml
  • pnpm-workspace.yaml
  • pyproject.toml
  • README.md
  • REVIEW.md
  • SECURITY.md
  • tsconfig.json
  • tsconfig.strict.json
  • uv.lock
  • zero

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More other

HKUDS/
Vibe-Trading

"Vibe-Trading: Your Personal Trading Agent"

33k
tinyhumansai/
openhuman

OpenHuman is an open source personal AI for Mac, Windows and Linux — local-first memory, agent orchestration, and deep research.

40k

😎 Awesome lists about all kinds of interesting topics [NOTE: Pull requests are temporarily disabled until I have a chance to catch up with the existing ones]

505k

AIPOCH Open-Science is an open-source, local-first, model-agnostic AI research workbench for macOS, Windows, and Linux, with scientific agents, Python/R notebooks, data connectors, and reproducible provenance.

4k
eigent-ai/
eigent

Eigent: The Open Source Cowork Desktop - Local and Free Alternative to Claude Cowork and Codex

15k