About bpm
bpm is a behavioral package manager for AI agents. It manages the conventions, extensions, templates, and measurement tools that shape how your agent works - the behavioral layer that sits between raw tools and the work you do together.
The way npm manages JavaScript dependencies, bpm manages the behavioral packages that make your agent effective. Install a convention and your agent shifts posture. Install an extension and your agent gains new capabilities. Install a template and your next project starts with structure.
Why bpm exists
Practitioners building with AI agents have discovered that the most impactful configurations aren't tools - they're behavioral patterns. How the agent approaches a task, when it delegates vs. scaffolds, what boundaries it respects, what posture it takes. These patterns are currently shared informally: copy-pasted between AGENTS.md files, referenced in blog posts, or locked inside individual workflows.
bpm makes these patterns installable, composable, and shareable. A team can publish their code review convention. A practitioner can share their learning mode protocol. A measurement tool can be installed alongside the convention it tracks.
The goal is an ecosystem where behavioral best practices compound across the community rather than being rediscovered in isolation.
How bpm differs from Agent Plugins
The Agent Plugins specification (associated with the Agentic AI Foundation) defines a standard for packaging reusable agent components - primarily skills and MCP servers. It answers the question: What tools does the agent have?
bpm answers a different question: How does the agent behave?
They're complementary layers. Agent Plugins packages the instruments. bpm packages the sheet music. A project can use both - Agent Plugins for tool distribution and bpm for behavioral configuration.
Agent Plugins
- Packages skills and MCP servers
- Defines tool capabilities
- Answers: what can the agent do?
- Standardized directory structure
bpm
- Packages conventions, extensions, templates
- Defines behavioral configuration
- Answers: how does the agent work?
- Registry + CLI + AGENTS.md integration
How bpm relates to AGENTS.md
AGENTS.md is a file that tells AI coding agents how to work on a specific project: build commands, test steps, code style, PR conventions. It's project-specific operational context - tied to a single repo.
bpm packages are portable behavioral patterns that travel with the practitioner across repos, projects, and teams. They define how the agent approaches work - not the build steps for one codebase, but the posture, conventions, and domain expertise that apply everywhere.
They work together: bpm packages define portable behaviors, and bpm generate writes a managed section into your AGENTS.md so the agent knows what's active. Your AGENTS.md contains both hand-written project context and bpm-managed behavioral packages.
AGENTS.md
- Project-specific operational context
- Build commands, test steps, code style
- Tied to one repo
- Answers: how does this codebase work?
bpm packages
- Portable behavioral patterns
- Posture, conventions, domain expertise
- Travel across repos and teams
- Answers: how should the agent behave?
Design principles
Packages live on GitHub
bpm is a registry and directory, not a host. Your code stays in your repo. bpm creates the connection between users and your work.
Context cost is visible
Every convention reports its token cost. Users can see exactly how much context window their installed packages consume and make informed decisions.
Agent-agnostic
bpm works with any agent that reads AGENTS.md - Goose, Claude Code, Codex, Cursor. The generated configuration section is standard markdown.
Open to all
Anyone with a GitHub account can publish. Names are first-come-first-served. The registry is open and the CLI is Apache-2.0 licensed.
Built by
bpm is created by Dakota Fabro, an engineer and researcher focused on human-AI collaboration, agent behavioral configuration, and the measurement of AI-assisted work.