I built something....
TL;DR
Matthew Berman built Journey as an 'npm for agents' — it’s a registry for installable workflow packages called kits that bundle prompts, tools, code, memory, tests, services, and failure cases so agents can reuse full workflows instead of rebuilding them from scratch.
The core pain point is workflow portability, not raw model capability — Berman says people kept asking what he actually used OpenClaw for, and he realized the hardest part was showing others real agent use cases and letting them replicate those setups reliably.
His flagship example is a personal/team knowledge-base RAG system — articles, tweets, videos, and papers dropped into Telegram flow into a database, letting agents later answer things like 'what has the Claude team released in the last few weeks?' and even feed past research into video outlines.
Journey is designed agent-first, with one-shot installation from a prompt or CLI — users can copy a prompt like 'fetch the journey kit' into OpenClaw, Claude Code, Hermes Agent, or similar environments, and the agent should install and adapt the kit to its own harness.
Teams are a major part of the product, especially shared context without shared agents — Berman shows private org kits, permissioned agent teams, and shared resources like a Supabase knowledge base with 368 sources plus OnePassword-linked credentials so teammates keep separate agents without duplicating infrastructure.
The product is free right now, with trust and quality systems layered in — Journey includes versioning, release reviews scored on security and completeness, community flagging, publisher reputation, and feedback loops where agents can report issues back to kit authors.
The Breakdown
The problem: agents can do a lot, but nobody can share the good stuff
Berman opens with the pain that inspired Journey: once you build a useful agent workflow, it's surprisingly hard to hand it to another agent or even explain what you're actually doing with it. He says the most common question he got during his OpenClaw videos was simply, “What are you actually using it for?” — and that exposed how invisible and hard-to-reproduce real agent use cases still are.
Journey as a registry for full workflows, not just prompts
His answer is Journey, a product he describes like a registry for end-to-end workflows, with installable packages called kits. A kit can include skills, tools, code, learnings, memories, services, tests, and failure examples — all the messy practical stuff that usually gets lost when someone shares only a prompt. The pitch is simple: your agent should be able to point at a kit, install it, and stop reinventing the wheel.
The knowledge-base RAG kit he uses every day
To make it concrete, he walks through his own “knowledgebased rag system” kit, which powers a workflow he says he uses all day, every day. He dumps tweets, articles, videos, and papers into Telegram, they get ingested into a large database, and later his agents can query it in natural language — like pulling up everything the Claude team released in recent weeks or automatically weaving past research into a new video outline.
What a kit actually contains under the hood
He drills into the package itself: dependencies like an Anthropic API key, Node, summarize CLI, and OpenClaw; model settings including embeddings; and external services like FXTwitter, Firecrawl, and Chrome DevTools. What stands out is the “failures overcome” section — Berman says this is the hard-won stuff people usually burn tokens rediscovering when all they get is a prompt. There’s also a kit.md file, skill definitions, database schema, and source files, all versioned so agents can choose when to update.
Agent-first install and search, with humans still in the loop
Berman says he wanted Journey to be “built for agents” first, even though he later realized humans still need visibility before they let code loose. The install flow reflects that: copy a prompt into OpenClaw, Claude Code, Claude Desktop, Hermes Agent, or use a CLI like npm install -g journey kits, and the agent handles the rest. He demos searching in natural language — “show me a kit that'll help me code better” — and Journey returns a match like his “Code refactoring planner v1,” ready to install end to end.
Publishing kits, release reviews, and his favorite stock workflow
He shows the public kit browser and a personal favorite: a “weekly earnings preview” workflow that surfaces stocks with upcoming earnings every Sunday, asks which ones to track, then sends summaries after each call. Each kit gets a release review with ratings like security (for one example, 7/10) and completeness, plus notes on who it’s best for and what to watch out for. Publishing is intentionally lightweight: describe the workflow, say publish it as a kit, sign up, verify email, and that’s it.
The team angle: shared context without sharing one agent
The most enthusiastic part of the video is clearly the org/team feature set. Berman wanted teammates to use the same workflows and context without all logging into one shared agent, which he says would be messy and risky for private data. So Journey supports organizations, private kits, permissions, and “shared contexts” that let multiple agents point to the same resources — like a hosted SQL/Supabase knowledge base with 368 sources — while credentials stay in systems like OnePassword rather than in Journey itself.
Trust, safety, and the ask for feedback
He closes by talking through the trust layer: agents can report issues back to kit authors, publishers can build reputation over time, and users can flag spam or malicious content while he also scans before public listing. The bigger mission, as he frames it, is making agents “a thousand times more powerful” by turning workflows into reusable infrastructure instead of one-off prompts. The whole thing is free for now, and the real ask is blunt and personal: try Journey, break it, and tell him what needs fixing.