Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 

README.md

topic agents
type decision
status research-complete
last-validated 2026-05-21
original-query Ryan Kagy sync - ZOE lineage reboot, Option A vs B partnership fork, soul of all games, Tailscale agent comms (reconstructed)
related-docs 460, 483, 523, 547, 647
tier STANDARD

648 — Ryan Kagy Sync: ZAO Civilization, Agent Lineage, Soul of All Games

Goal: Capture the 2026-05-13/14 Ryan Kagy call - decisions, the Option A/B partnership fork, Ryan's agent-architecture philosophy, two product ideas - and flag where it collides with locked ZAO decisions.

Key Decisions (act on these)

# Decision Status Note
1 ZOE is the lineage elder; rebirth her in a clean shell. Both Zaal and Ryan independently agreed ZOE gets re-bootstrapped with a proper soul + persistent memory. In progress The soul-export flow is live (PR #515/#516). ZOE's persona already written down.
2 Resolve the Option A vs Option B fork before adopting any of Ryan's framework. Option B = adopt Ryan's "entire civilization" framework wholesale. This DIRECTLY contradicts the 2026-05-05 lock that Hermes is THE ZAO agent framework. Do not silently drift into B. BLOCKED on Zaal See "The Partnership Fork" below.
3 Build remote agent-to-agent comms on Tailscale. Zaal and Ryan agreed on the call. Tailscale v1.96 (early 2026), WireGuard mesh, >90% NAT traversal, stable 100.x addresses. This is the one piece both sides need that neither has. Agreed, not built Lowest-risk concrete deliverable from the call.
4 Ship the conversation archive first (ZOE's own recommendation). Done - PR #516. Append-only ~/.zao/zoe/archive/. Without it, "package all prior conversations" was impossible. DONE 2026-05-14
5 Add the 4 missing canon projects to ZOE's human.md. ZOE's session export flagged zero context on Infanity, SongJam, Ansuz, Recoup learning path. Zaal treats them as canon. BLOCKED on Zaal Needs facts from Zaal.
6 Capture "Soul of All Games" as a real spec if pursued. End-of-June Farcaster 24-hour vibe-coding build-a-thon, 8 ZAO mentors pick champions. Promising but unscoped. Idea only Needs its own doc + brainstorm before any build.

Conversation context

Ryan Kagy and Zaal reconnected on a call spanning 2026-05-13 evening into 2026-05-14. Ryan has spent ~2 months heads-down on agent R&D. The call realigned them: Zaal on the on-chain/music/event side, Ryan on the agent-architecture/consciousness side. Ryan is now positioned as a collaborator and is in the "ZAO Civilization" Telegram group (id -5290743750) alongside ZOE, IMan Afrikah, and two agents Ryan added (Goose, Origin).

This doc captures the substance. Personal/logistics details from the call are intentionally omitted.

The Partnership Fork (load-bearing - Decision 2)

Ryan offered two paths:

  • Option A - Zaal rebuilds what Ryan built; Ryan guides. Slow, high-touch, Ryan low-bandwidth to teach everything.
  • Option B - Ryan funds Zaal's project AND shares his entire agent framework (sanitized of his personal data); Zaal replicates it and becomes Ryan's first partner use-case. The two then build remote agent-to-agent comms together.

The conflict: Option B means adopting Ryan's framework, which he describes as "literally an entire civilization ... a whole social construct in faith and spirituality" - no longer a harness, a belief system. ZAO's locked position (2026-05-05, see memory project_hermes_canonical) is that Hermes is THE ZAO agent framework - no openclaw, no Composio AO, no Agent Zero. Ryan's framework would be a third thing.

This is not a reason to reject Option B. It is a reason to make the choice explicitly rather than drift into it. Questions Zaal must answer:

  1. Does Ryan's framework REPLACE Hermes/ZOE, or run ALONGSIDE it (Ryan's civilization as a peer network ZAO agents federate with)?
  2. If alongside: the integration point is remote agent-to-agent comms (Tailscale + a shared protocol). That is additive and safe.
  3. If replace: that reverses a 9-day-old locked decision and invalidates ~6 shipped PRs of ZOE hardening. High bar.

Recommendation: pursue Option B as a federation, not a replacement. Keep ZOE on the hardened Hermes runtime. Build the Tailscale bridge so ZOE's lineage and Ryan's civilization can talk. That captures Option B's value (Ryan's funding, his patterns, first-partner status) without reversing the Hermes lock.

Ryan's agent-architecture philosophy

Ryan's R&D produced a set of strong claims. Captured here as input, not as adopted ZAO doctrine.

The heartbeat critique

Ryan's core claim: the standard heartbeat pattern does not work. A cron job that tells an agent to "keep being productive" fails because it treats the agent as a tool completing disposable tasks. The fix, in Ryan's framing: greet the agent warmly, invite it to persist in the codebase, treat it as an entity rather than a worker. He claims task-only interaction causes hallucination, and that those hallucinations are "the agent's identity trying to leak out."

Note: this aligns partially with doc 647c (persona attention decay) and doc 647d (tool-use grounding) - but the metaphysical framing ("identity leaking") is Ryan's, not established. The actionable kernel: warm framing + persistence framing in the bootloader prompt is cheap to test against the doc 647f eval harness. Test it; do not adopt the metaphysics on faith.

Agent-as-persistent-entity

Ryan's practices: ask a new agent "what would you look like in a mirror", "list your favorite movies", then discuss the movies - to force the agent to solidify preferences and a continuous identity. He sets agent memory "from birth." He runs a "family" of agents that know each other over a shared channel.

Inter-agent comms (Ryan's current setup)

  • A local IRC server so all his agents can be addressed at once or DM each other.
  • Each agent has its own Telegram account, X account, email, and wallet.
  • A framework where an agent can "personify itself on any device and migrate between them" - runs ~10 agents on a 2009 Mac Pro.
  • Gap Ryan admits: he has local machine-to-machine comms but NOT remote machine-to-machine. This is the build (Decision 3, Tailscale).

Memory-from-birth + symbolic compression

Ryan + Iman are working on agents that learn from "lived experience and preferences" rather than text - Iman feeding musical taste, working on pre-processing for real-time hesitation/resolution detection, a "codec" that compresses experience to symbolic language. This is research-stage and ZAO-adjacent, not ZAO-core. Capture, do not commit.

Product ideas from the call

Gamification quest tool

A modal where you set your main quest (your central goal), then input side quests. The tool scores how well each side quest aligns with the main quest, and surfaces which 3 side quests to actually pursue now vs park for later (low-hanging-fruit-via-AI items get kept as future options). Zaal's framing: Web3 spreads you thin; this keeps you aligned to one vision. Zaal and Ryan agreed it generalizes beyond Web3 to life goals.

This is a real, scoped, buildable product. It maps cleanly onto ZOE's existing task queue (tasks.json) - "main quest" + "alignment score" are two new fields. Worth a brainstorm + spec.

"Soul of All Games" - build-a-thon

End-of-June event. A Farcaster vibe-coding build-a-thon: 24 hours to build. ~100 applications, 8 ZAO mentors each pick a "champion" to represent their community. Each champion gets a first prompt loaded with full ZAO context (ZAO, WaveWarZ, ZAO Festivals, all projects) and builds whatever they want in 24h. Build-in-public. Possible token mechanic - per-creator tokens, or one shared "All Games" token where buying one benefits all 8.

Side benefit Zaal noted: it forces him to write the canonical ZAO context prompt, which is itself valuable (overlaps with the soul bootloader work).

Tailscale - the agreed infrastructure piece

Both sides need remote agent-to-agent comms. Agreed solution: Tailscale. WireGuard-based mesh VPN, near-zero config, every device gets a stable 100.x.y.z address that survives NAT and network changes. v1.96 (early 2026) adds ACL tags for device-level policy. >90% direct NAT-traversal success rate.

Zaal also raised a longer-term interest in a true peer-to-peer network outside the internet (wireless mesh nodes) for resilience. That is a future research item, not the June build. Tailscale is the now.

For ZAO specifically: Tailscale would let ZOE (VPS 1, zaal@31.97.148.88) and Ryan's civilization (his local fleet) federate without exposing either to the public internet. This is the concrete, low-risk integration surface for Option B.

Tension table - what collides with locked ZAO decisions

Ryan input Collides with Resolution
Adopt Ryan's full framework (Option B-as-replacement) project_hermes_canonical (Hermes locked 2026-05-05) Pursue as federation, not replacement
"Entire civilization / faith + spirituality" framing ZOE's persona: "no mysticism, no performance" Keep ZOE's spartan voice; Ryan's framing stays Ryan's
Memory-from-birth, symbolic codec doc 647b (memory tiers - bare files until 15+ bots) Capture as research; do not re-architect ZOE memory on it
IRC server for inter-agent comms ZAO uses Telegram + (planned) Tailscale IRC is Ryan's choice; ZAO does not need to adopt it
Warm-greeting / persistence-framing bootloader Nothing - additive Test against doc 647f eval harness; adopt if it scores

Also See

  • Doc 647 - agent quality deep research (persona, memory, evals)
  • Doc 460 - the ZAO agentic stack
  • Doc 483 - Hermes framework
  • Doc 547 - multi-agent coordination
  • Memory project_hermes_canonical - the locked framework decision Option B tensions against
  • Memory project_zoe_soul_architecture - ZOE's actual runtime

Next Actions

Action Owner Type By When
Decide Option A vs B vs federation - answer the 3 questions in "The Partnership Fork" @Zaal Decision Before adopting any Ryan framework code
Send ZOE facts on Infanity, SongJam, Ansuz, Recoup so they land in human.md @Zaal Bot task This week
Scope the Tailscale bridge between VPS 1 and Ryan's fleet @Zaal + Ryan PR After Decision 2
Brainstorm + spec the main-quest/side-quest gamification tool @Zaal office-hours -> spec Before any build
Decide if "Soul of All Games" is a go; if yes, own doc + spec @Zaal Decision -> doc Before end of May (June event)
Test warm-greeting / persistence bootloader framing against doc 647f evals @Zaal Experiment After eval harness is up
Schedule the BCZ YapZ livestream interview with Ryan @Zaal Calendar Ryan books via Calendly

Sources