# The code repository behind a multi-agent book, and the four places it disagrees with itself > victordibia/designing-multiagent-systems ships PicoAgents plus fourteen example directories to go with Designing Multi-Agent Systems. Two chapters promise no code, the chapter table stops mid-cell, and a premium-samples/ folder sits at the root. **victordibia/designing-multiagent-systems** — Building LLM-Enabled Multi Agent Applications from Scratch - Repository: https://github.com/victordibia/designing-multiagent-systems - Website: https://buy.multiagentbook.com/ - Stars: 1,337 · Forks: 276 - Language: Python - License: Apache-2.0 - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/victordibia-designing-multiagent-systems ## The repository description and the book title are different sentences The one-line repository description reads Building LLM-Enabled Multi Agent Applications from Scratch. The README names something else: the official code repository for Designing Multi-Agent Systems: Principles, Patterns, and Implementation for AI Agents by Victor Dibia. From Scratch and Principles, Patterns, and Implementation are different pitches, and only the second one matches the chapter tables you actually find inside. The repository is Python, Apache-2.0, on the main branch, with a LICENSE at the root and 1334 stars, 275 forks, and 7 open issues. It was last pushed to on 2026-08-24. The homepage field points at buy.multiagentbook.com rather than at documentation, which tells you the repository is positioned as the companion to a sale. What the repository actually contains is a teaching framework named PicoAgents, described as a full-featured multi-agent framework built entirely from scratch for the sole purpose of teaching you how multi-agent systems work. That word sole is doing real work in the pitch. This is not positioned as a framework you adopt in production. It is positioned as the thing you read once to understand agent reasoning loops and orchestration patterns, and the root listing is arranged to make that legible: picoagents/ for the framework, examples/ for the runnable material, docs/, images/, and course/, code_along/, research/, premium-samples/ for everything around it. ## Chapters 2 and 3 carry a dash where the code should be The book is organized across 4 parts, taking you from theory to production. Part I, Foundations of Multi-Agent Systems, maps three chapters onto a table with a Code column and a Learning Outcome column. Chapter 1, Understanding Multi-Agent Systems, points at a poet/critic example under examples/workflows/yc_analysis/, with the learning outcome of understanding when multi-agent systems are needed. That example is a YC company analysis workflow, which tells you the first thing a reader builds is an analysis pipeline rather than a chat bot. Chapters 2 and 3 put a plain dash in the Code column. Chapter 2 is Multi-Agent Patterns, aimed at mastering coordination strategies, workflows versus autonomous. Chapter 3 is UX Design Principles for Multi-Agent Systems. Neither ships a file reference in this repository. So of the first three chapters, the one whose outcome is a pattern vocabulary has no code, and the one whose outcome is interface design has no code. For a repository whose stated differentiator is that every theoretical concept is backed by working, tested code, the two dashes are the first place to look before you plan study time around them. ## The Chapter 6 row ends mid-cell in the Learning Outcome column The Part II table for Building Multi-Agent Systems from Scratch has four rows in the layout and does not finish them. Chapter 4, Building Your First Agent, is the widest cell in the document: agents/_agent.py, then basic-agent.py, memory.py, middleware.py, structured-output.py, agent_as_tool.py, then otel/, memory/, and tools/approval_example.py, and finally a link to a Colab notebook at examples/notebooks/01_basic_agent.ipynb. Its outcome reads build agents with tools, memory, streaming, middleware, observability, and human-in-the-loop. Six capabilities, nine paths, one cell. Chapter 5, Computer Use Agents, points at agents/_computer_use/ and examples/agents/computer_use.py for browser automation agents with multimodal reasoning. Chapter 6, Building Multi-Agent Workflows, points at workflow/ inside picoagents/src/picoagents/ and at examples/workflows/, and its Learning Outcome cell starts Build type-safe workflows with streaming and stops there. So the table ends inside Chapter 6's outcome text, with the description cut off mid-phrase. That leaves the Chapter 6 outcome, anything the book covers after Chapter 6, and Parts III and IV entirely undescribed in the part of the page shown here, even though the book is said to run across four parts. The Colab notebook appears on exactly one row, so browser-only execution is offered for a single chapter while the rest of the material is expected to run from a clone. ## premium-samples/ sits at the root next to picoagents/ The root listing runs: .devcontainer/, .gitignore, .run_samples.sh, Citation.cff, LICENSE, README.md, code_along/, course/, docs/, examples/, images/, picoagents/, premium-samples/, research/. Three of those directories name a commercial tier or a body of teaching material: code_along/, course/, and premium-samples/. The naming tells you the repository is not one thing. PicoAgents is the Apache-2.0 framework. course/ and code_along/ imply structured teaching material that may or may not follow the book. premium-samples/ names a paid layer, and it sits at the same depth as the free framework, so whether a given sample is reachable depends on a directory boundary rather than a documented rule. The README shown here does not explain what separates those directories, and there is no file listing at root that maps them onto chapters. Two files next to them are more useful than they look. Citation.cff is a real citation metadata file, which means the project is set up to be cited mechanically. .run_samples.sh is a shell script at the root, which is the closest thing to a one-command entry point in the whole listing. Neither is mentioned in the chapter tables. ## examples/frameworks/ in a repository that sells framework-agnosticism One of the four stated differentiators is Framework-agnostic: core patterns that transcend any specific framework, with the parenthetical reason that it avoids the lock in or outdated api issue common with books that focus on a single framework. Another is Fundamentals-first, building from scratch to understand every component and design decision. A third is Production considerations covering evaluation, optimization, and deployment guidance from real-world experience. The examples/ directory then holds fourteen subdirectories: agents/, app/, contextengineering/, evaluation/, frameworks/, mcp/, memory/, notebooks/, optimization/, orchestration/, otel/, tools/, webui/, and workflows/, next to its own README.md. Several of those map directly onto the differentiator list: evaluation/ and optimization/ for Production considerations, otel/ for the observability promised in Chapter 4, mcp/ and webui/ for the protocol and interface layers the chapter tables never mention. The one that pulls against the pitch is frameworks/. A repository that argues against teaching a single framework still carries a directory named for frameworks. The directory name alone does not say whether it compares them, wraps them, or benchmarks them, and the chapter tables do not reference it. What it does say is that the framework-agnostic claim is a claim about PicoAgents' core patterns rather than about the repository's entire surface. ## Release notes that fit on one line, with the last tag six months behind the last commit Three releases are published. v0.4.0 from 2026-02-11 is titled Hooks, Compaction, Context Tools, Benchmarks. v0.3.2 from 2025-12-28 is titled Fix AgentResponse print output. v0.3.1 from 2025-11-23 carries only its own version number as a title. Each title is a single line, and two of the three describe either a print-formatting fix or nothing at all. There is no changelog file in the root listing and no per-chapter mapping from a release to the examples it changed, so a reader who wants to know whether a specific sample changed has no document to consult and has to compare commits instead. The version shape is also worth noticing. v0.4.0 bundles four unrelated capabilities, hooks, compaction, context tools, and benchmarks, under one minor bump. That is a project treating the framework as a teaching surface where new patterns get added together rather than shipped one at a time, which suits reading order and makes release-to-release diffing a weak signal. The timing gap matters more: the newest tag is dated 2026-02-11 and the newest commit is dated 2026-08-24. Anything in the tree that is newer than v0.4.0 is unreleased as far as a pip-style consumer is concerned. ## Two empty link badges and three purchase links carrying tracking parameters The top of the README carries two image links with empty link text, one pointing at codespaces.new/victordibia/designing-multiagent-systems?quickstart=1 and one at buy.multiagentbook.com. With no alt text and no caption, a rendered page gives you two unlabelled targets, and the first of them is the fastest route to a working environment. The purchase links do carry labels: Buy Digital Edition, Paperback on Amazon, and Hardcover on Amazon. All three resolve to the same destination pattern with utm_source=github&utm_medium=readme appended, so the repository is the marketing surface for the store rather than a neutral mirror of it. Between the two groups, the README never states how to obtain the code. There is no install line, no package name to install, and no version requirement in the text above the Chapter 6 row. What exists instead is structure: a .devcontainer/ directory for a local environment, a Codespaces quickstart link for a hosted one, and a root script called .run_samples.sh. A reader is expected to derive the setup from the directory layout rather than from a documented command. ## Conclusion This repository earns its place if you want the framework-agnostic patterns from Designing Multi-Agent Systems as code you can read end to end, and you are willing to treat the chapter tables as a rough map rather than a syllabus. It does not serve anyone looking for a chapter-by-chapter curriculum, a fully worked multi-agent codebase, or evidence of what changed recently. Before you plan around it, check which chapters actually ship code, read the root listing to see which directories sit outside the free tier, and date your expectations from the v0.4.0 tag in February 2026 rather than from the last commit. ## FAQ ### What framework does the designing-multiagent-systems repository ship? PicoAgents, described as a full-featured multi-agent framework built entirely from scratch for teaching multi-agent systems rather than for production use. The chapter tables link into picoagents/src/picoagents/, including agents/_agent.py, agents/_computer_use/, and workflow/. ### How do I run the examples in designing-multiagent-systems? The root holds a .run_samples.sh script and a .devcontainer/ directory, and the README offers a one-click Codespaces link to the repository with a quickstart parameter. No install or pip command appears in the text above the Chapter 6 row of the chapter table. ### Which chapters of Designing Multi-Agent Systems come with code? In Part I only Chapter 1 lists code, a poet/critic example under examples/workflows/yc_analysis/. Chapters 2 and 3 carry a dash in the Code column. Part II lists paths for Chapters 4, 5, and 6, and the table ends partway through Chapter 6's learning outcome. ### Is the designing-multiagent-systems repository free to use? The repository is licensed Apache-2.0 with a LICENSE file at the root. The README also links a Digital Edition, a Paperback, and a Hardcover, and the root listing includes a premium-samples/ directory alongside code_along/ and course/. ## Sources - [License: Apache-2.0](https://github.com/victordibia/designing-multiagent-systems/blob/main/LICENSE) - [Project website](https://buy.multiagentbook.com/) - [README](https://github.com/victordibia/designing-multiagent-systems/blob/main/README.md) - [Releases](https://github.com/victordibia/designing-multiagent-systems/releases) - [victordibia/designing-multiagent-systems on GitHub](https://github.com/victordibia/designing-multiagent-systems) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/victordibia-designing-multiagent-systems