This repository holds the [&] Protocol specification and every in-tree reference implementation of it, plus the portfolio products that build on top. This document is the one-page tour for someone trying to understand how the pieces fit.
For authoritative status — what's shipped, what's in progress, what the test counts are — see `STACK_COMPLETION.md`.
┌──────────────────────────────────────┐
│ [&] Protocol — composition algebra │
│ 6 primitives · 2 operators │
│ (AmpersandBoxDesign/) │
└───────────────┬──────────────────────┘
│ compiles into
┌───────────────────┴────────────────────┐
│ │
┌───────┴────────┐ ┌────────┴────────┐
│ MCP runtime │ │ A2A coordination│
│ transport │ │ transport │
└───────┬────────┘ └─────────────────┘
│
┌──────────────┼───────────────────────────────────────────┐
│ │ │
│ OS protocols (OpenSentience) — 11 normative protocols │
│ OS-001 Continual Learning → Graphonomous │
│ OS-002 κ-Routing → Graphonomous │
│ OS-003 Deliberation → Graphonomous │
│ OS-004 Attention → Graphonomous │
│ OS-005 Model Tier → (system-level) │
│ OS-006 Governance Shim → delegatic │
│ OS-007 Adversarial → (draft) │
│ OS-008 Agent Harness → open_sentience │
│ OS-009 PRISM benchmark → PRISM │
│ OS-010 PULSE temporal → PULSE │
│ OS-011 Embodiment → body-browser + body-os │
└──────────────────────────────────────────────────────────┘
The north-star use case for this stack is the dark factory: agents hosted on machines that operate autonomously, learn continually, and transfer skills to each other. Seven steps close the loop:
┌──────────────────────────────────────────────────────────────┐
│ 1. Machine A perceives & acts ← body-browser │
│ body-os │
│ 2. Records InteractionTrace ← Graphonomous │
│ (&memory.episodic.store) │
│ │
│ 3. Learns via memory consolidation ← Graphonomous │
│ (consolidation, κ-routing) │
│ │
│ 4. Crystallizes to SkillCandidate ← FleetPrompt │
│ (Crystallizer + Oban PollWorker + │
│ GraphonomousClient.HTTP) │
│ │
│ 5. Ships via FleetPrompt ← FleetPrompt │
│ (Registry, Trust, InstallEngine) │
│ │
│ 6. Machine B installs + replays ← body-browser │
│ (with fresh authorization on body-os │
│ destructive edges per OS-011 §4.4) delegatic │
│ (authorizer) │
│ │
│ 7. PRISM measures ← PRISM │
│ (OS-011 scenario suite, │
│ BYOR cross-system benchmark) │
└──────────────────────────────────────────────────────────────┘
Current status (see STACK_COMPLETION.md for always-live detail): 7 of 7 steps fully in-tree. Every step has implemented code paths, passing tests, and (for the MCP-exposed pieces) deployment configs.
| Product | Role in the loop | Transport | Test count |
|---|---|---|---|
| Graphonomous | episodic memory + continual learning engine | MCP (stdio/HTTP) | 573 |
| PULSE | temporal loop manifest standard | npm package | 3 |
| PRISM | cross-system CL benchmark | MCP + LiveView | 172 |
| [&] reference | protocol spec + Elixir CLI | CLI | 58 |
| body-browser | &body.browser provider (OS-011) | MCP (stdio/HTTP) | 41 |
| body-os | &body.os provider (OS-011) | MCP (stdio/HTTP) | 41 |
| delegatic | OS-006 authorization kernel | MCP (stdio/HTTP) | 42 |
| Product | Role | Tests |
|---|---|---|
| Agentelic | spec-driven agent builder | 66 |
| FleetPrompt | marketplace + trust + install + crystallizer | 121 (108 + 13 install tests) |
| SpecPrompt | spec-driven dev toolchain | 93 |
| RuneFort | spatial cognition visualization (SvelteKit) | — |
WebHost.Systems (React dashboard, 143 tests) and BendScript (SvelteKit KAG editor, 26 tests).
AgenTroMatic (1118-line spec), Delegatic (630), Deliberatic (418), GeoFleetic (987), TickTickClock (1013).
All three fit together as a three-layer stack:
[&] (composition) — agents declare what capabilities they have via .ampersand.json.
PULSE (temporal) — agents declare how their capabilities circulate via .pulse.json.
PRISM (diagnostic) — benchmarks measure the loops over time.
Example: Graphonomous declares its continual-learning loop in PULSE/manifests/graphonomous.continual_learning.json; PRISM reads that manifest directly to discover phases, signatures, and substrates — no separate benchmark config needed.
Supabase: single shared instance, fleet.*/kag.*/webhost.*/…/spec.* per-product schemas. One supabase start runs every product's data layer. 35 migrations live in ampersand-supabase/migrations/.
amp-nav: Web Component distributed to every portfolio marketing site via sync-nav.sh. Current version: v0.3.3.
ampersand-plugins: Claude Code skill packs (graphonomous, prism, ampersand-protocol, spec-driven-dev).
Fly.io deployments: three live today (graphonomous-mcp, prism-eval, os-pulse-mcp); body-browser-mcp and body-os-mcp have production fly.toml + Dockerfile ready to deploy.
Each product's in-repo documentation entry point. These links resolve on GitHub and in the docs site:
[&] Protocol (composition) — docs hub · architecture · the umbrella pipeline
The three-protocol stack — [[&] · PULSE · PRISM, working together](PULSE/docs/THREE_PROTOCOL_STACK.md)
PULSE (temporal) — OS-010 specification
PRISM (diagnostic) — OS-009 specification
OpenSentience (OS protocols) — docs hub · spec index
BendScript (documents) — protocol spec
WebHost.Systems (runtime) — docs hub · master spec
Agentelic — docs hub · AgenTroMatic — docs hub · Delegatic — docs hub · Deliberatic — docs hub
FleetPrompt — docs hub · SpecPrompt — docs hub · GeoFleetic — docs hub · TickTickClock — docs hub
RuneFort — protocol spec · Workbench — spec · Weave — design
Cross-cutting docs at the repo root that frame the whole portfolio:
Repository README — top-level entry point and quick orientation
Agent guide (AGENTS.md) — how agents should operate across this monorepo
Stack status — authoritative per-product versions, test counts, deployment state
Stack planning — roadmap and sequencing across products
Stack architecture gap review — known gaps between spec and implementation
Stack sitemap — every product, domain, and doc surface in one map
Documentation atlas — the filesystem-mirror docs site that renders this corpus
Spec writer: start with AmpersandBoxDesign/SPEC.md and opensentience.org/docs/spec/
Memory-system builder: start with graphonomous/docs/spec/README.md and graphonomous/CLAUDE.md
Body provider builder: start with body-browser/README.md, then body-os/README.md
Benchmark user: start with PRISM/docs/ and PRISM/priv/fixtures/os011/README.md
Marketplace developer: start with fleetprompt.com/docs/spec/README.md
Everything at once: this file + STACK_COMPLETION.md + CLAUDE.md at the repo root
Everything in-tree is Apache-2.0 unless noted otherwise in the directory's LICENSE file.