Sovereign Agent Resources

Tutorials, example projects, and reference code for practitioners building governed AI organizations with Sovereign Agent and ZeoCore.

Every resource is a directory you can run in full, study, or copy piece by piece into your own project. Each one is pinned to versions it was verified against, and the pins are kept honest by machinery, not by promises.

---

Start here

New to Sovereign Agent. No account, no API key, no license — the whole loop runs offline:

uvx sovereign-agent@latest doctor
uvx sovereign-agent@latest demo store --mode simulated --root /tmp/first-shift

Then go deeper, in order:

1. The book — thirteen chapters that build a Zero-Employee Organization from an empty directory, break it on purpose, and repair it. Every code block executes; every output is byte-verified.
2. examples/sovereign-agent-live-demo — a real local LLM tool-calls a ZeoCore capability and the Sovereign Agent governs the result. bash setup.sh, then three uv run commands.
3. examples/zeocore-examples — three real applications rebuilt on zeocore: typed tools doing actual work.

Building something specific. Open the category catalog that matches your goal: examples, tutorials, patterns.

What this is

• Example projects — complete, clone-and-run demonstrations, not fragments
• Tutorials — end-to-end walkthroughs, each self-contained and runnable
• Reference code — patterns for recurring problems: governed tools, verification, refusal as a first-class result
• Workshop material — slides, exercises, and solutions from live sessions

What this is not

• Not the framework. sovereign-agent and zeocore live on PyPI and in their own repositories; the executable textbook lives in profrodai/sovereign-agent. This repository is the material around them.
• Not a support channel. Product issues belong on the product repositories; questions about material here belong in Discussions.

---

Zeocore 0.10.0 catalog update

The release labs cover HubSpot and Kit marketing, managed test/production state, native service resolution, Gemini request construction, and local conversion/execution receipts. Read the integration account tracks for exact key acquisition, test-account and production-account instructions before connecting real services. The catalog also includes the governed Notion and bounded Supabase resources.

The direct Zeocore pins now use the published 0.10.0 release. Sovereign Agent remains at 1.4.0. The ZEO Creator handoff retains its documented immutable Git dependency and transitive Zeocore 0.9.0; it is revalidated in that environment.

Featured resources

Resource	What you watch happen	Path
Six advanced patterns	Isolation, scheduling, context, fencing, discovery, and memory fail safely in six offline labs	examples/sovereign-agent-advanced-patterns
Live governance demo	A real local model proposes a restock; deterministic Python re-validates it against the ledger and refuses 9999 units	examples/sovereign-agent-live-demo
Data cleaning on zeocore	B2B contact/company CSV cleaning as typed tools	examples/zeocore-examples/apps/data_cleaning
Doc → Bluesky pipeline	A document becomes governed social posts	examples/zeocore-examples/apps/doc_to_bluesky
Metrics tracker	Governed metric collection on zeocore	examples/zeocore-examples/apps/metrics_tracker

---

Repository map

The repository holds two kinds of material under two different promises.

Maintained catalog — one shared pin per governed package (SOVEREIGN_AGENT_VERSION, ZEOCORE_VERSION), migrated together, expected to stay green for as long as it is checked in.

Path	What it holds
examples/	Complete clone-and-run projects
tutorials/	Step-by-step walkthroughs — accepting contributions
patterns/	Small reference implementations of recurring problems — accepting contributions
snippets/	Short pieces too small to be a pattern — accepting contributions
workshops/	Slides, exercises, and solutions from sessions — accepting contributions

Frozen snapshots — author-pinned, never migrated forward, still required to be reproducible (a committed uv.lock, a name, a date). See docs/SNAPSHOTS.md.

Path	What it holds
community/	Resources written by practitioners, credited to their authors — accepting contributions

---

Staying current

The pins are governed by tooling, never by hand (docs/MIGRATING.md):

make outdated    # ask PyPI what is new — read-only
make update      # bump both pins to newest, rewrite every project, re-lock
make ci          # install every project for real and prove the pins resolved

make migrate VERSION=x refuses a version that does not exist on PyPI — before writing anything. make status fails if any project drifts from the root pins. CI runs the offline gate on every push and asks PyPI weekly whether the catalog has gone stale.

Requirements

• uv — supplies Python and every environment; nothing else to install
• Ollama — only for the live-model demos; everything else runs offline with the deterministic scripted provider
• No API keys, no accounts, no licenses for the core material

Contributing

Start with CONTRIBUTING.md; AI agents get the same contract as a deterministic procedure in AGENTS.md. The short version: make new-resource CATEGORY=patterns NAME=my-resource scaffolds a directory with a pinned pyproject.toml and the required metadata block — fill it with something a stranger can run, uv lock, and pass make validate.

License

MIT, unless a resource's own directory states otherwise.