Hash-verified installs
Each file is checked against the lockfile before it’s written. facet verify detects drift after installation.
Give every developer’s AI assistant the same approved skills and tools, kept in sync automatically.
Pilot openTeam tier free for 6 months for design partners →Who changed what, who approved it, and when. Every update is reviewed before it reaches anyone’s assistant.
Today, shared agent configuration lives in copied folders, per-tool directories, and chat threads. It drifts quietly until something breaks.
facet install restores the exact locked configuration and verifies every file by hash.One command finds a facet in the public registry, verifies every file by hash, and installs it for each coding tool your team uses.
$curl -fsSL https://agentfacets.io/install | bashPublic facets are free to download and publish, always. Sign up to publish →
Open beta · MIT-licensed CLI and open specification
The same building blocks your team already trusts for code: manifests, lockfiles, hashes, and review.
One locked facet works across every supported coding tool. Adapters are open source and straightforward to write, so you don’t wait for vendor support.
Write your own adapter with the SDK →Exact versions, pinned by hash. The manifest records intent and the lockfile records resolution, and both are version-controlled. Every environment resolves identically.
Read the open specification →Every file is hashed and verified before anything is written to disk, so what was published is exactly what you install. facet verify confirms installed state matches the lockfile at any time.
Facets that build on facets. Compose a front-end facet from the TypeScript and React facets you already trust, with one resolved graph and one lockfile.
See the roadmap →A facet puts text into an agent’s context and can declare servers that run commands. That’s why content verification and execution consent are part of the core specification, not optional hardening.
Each file is checked against the lockfile before it’s written. facet verify detects drift after installation.
Manifest and lockfile live in your repository. A new version or a new MCP server shows up as a diff your reviewers approve.
MCP servers are declared explicitly in the facet manifest, so you can see what a package wants to run before you install it.
MIT-licensed CLI, an open specification, and an install script you can read before running it. No black box between you and your agents.
Private facets are visible only inside your organization’s account. They are never published publicly or shared with other customers.
Security and platform admins pre-approve which MCP servers are allowed to run. Pilot partners are shaping this workflow now.
We’ll walk your security and platform teams through the integrity model, install behavior, and MCP server handling, and map them against your requirements.
Start with the team that maintains developer tooling. Give engineering leaders and security a concrete configuration to evaluate.
Scope a pilot ↗Maintain shared capabilities across tools and repositories from one source. Retire the hand-maintained sync scripts, symlinks, and copy steps.
Roll out agent configuration the way you roll out any dependency: versioned, reviewed, and reproducible, with a bounded pilot before you expand.
Review package contents, sources, and MCP server declarations before they reach a developer machine. See the trust model →
The CLI, specification, and public registry stay free. Teams pay for private distribution and organization controls.
For individual developers and open-source publishers.
For engineering teams standardizing agent configuration.
For organizations with security, procurement, and rollout requirements.
No. Developers keep the coding agents they already use. Agent Facets packages and installs capabilities for them, and adapters handle each tool’s native layout.
Yes. On the Team tier, private facets are visible only inside your organization’s account. Nothing you publish privately is exposed publicly or shared with other customers.
Claude Code, OpenCode, and Codex today. The open-source Adapter SDK lets you add any other harness without waiting for us.
The configuration: every packaged asset and resolved version, pinned by hash in facets.lock. Model output still varies with the model and runtime, but with an identical configuration, differences come from the model and not from configuration drift.
MCP servers are declared in the facet manifest and arrive in your repository as reviewable changes. An install-time approval workflow, where admins pre-approve what may run, is in active development. If your organization needs it before a wider rollout, tell us during the pilot. See the trust model →
A 30–45 minute kickoff call where we install together, live, in your real repository. Then 2–6 weeks of everyday use with day-3 and day-14 check-ins and a direct line to the founding team. The Team tier is free for 6 months, and there’s no obligation at the end.
Yes. The lockfile records exactly what was installed, so removal is clean and rolling back is a normal revert.
Bring one piece of configuration you already maintain by hand: a shared prompt, a house style guide, an internal MCP connection. We package it with you and install it in your real repository on the kickoff call.
Prefer to explore on your own? Read the documentation ↗
MIT-licensed CLI, adapters, and Adapter SDK.
StandardSpecification ↗The open facet format, lockfile, and integrity model.
ReferenceCLI reference ↗Every command, flag, and output.
MomentumChangelog ↗What shipped, release by release.
DirectionRoadmap ↗What’s next, including composition and install-time approval.