Part of the [&] Protocol ecosystem

Walk through,
the dark factory that
ships agents.

Other dark factories ship code. RuneFort ships agents — autonomous capabilities that keep running, learning, and producing output forever. Point it at a PULSE manifest. Your loop becomes a floor plan. Press R! and the spatial UI drives SpecPrompt → Agentelic → FleetPrompt → deploy — composed agents come out the far end.

Three steps. That's the product.

1

Drop a manifest

Point RuneFort at a *.pulse.json. It generates the fort — rooms from phases, corridors from signals, walls from policies.

2

Walk the fort

Zoom from district to rune. Hover to reveal flow, thermal, or diagnostic. No menu to learn — the right lens fades in where you look.

3

Run the factory

Press R! and the spatial UI drives Agentelic → FleetPrompt → WebHost — composed agents come out the far end.

The fort is what you understand. The runes are what you inspect.

— Manifests stay canonical. The view just makes them walkable.

Other dark factories ship code.
RuneFort is a Dark Agent Factory.

Code is a terminal artifact — you deploy it, it runs, that's the end. An agent is a generative artifact — you deploy it, and it keeps producing output forever. A Dark Agent Factory is a dark factory whose output is itself another dark factory. That's not a feature difference. It's a level up.

L0
Human writes code
One feature runs, once.
L1
AI assistant writes code
One feature runs, once. Faster.
L2
Dark factory writes code
One feature runs autonomously. Repeat per feature.
L3
Dark Agent Factory deploys agents
Every deploy ships another autonomous worker. Output compounds.

Six advantages of a Dark Agent Factory

Recursive leverage

A shipped PR runs once. A shipped agent ships things every time it wakes. One day of the RuneFort factory deploys N agents; those N agents produce work indefinitely. Code-factory output is additive. Agent-factory output is multiplicative.

Composability — [&] Protocol

Every shipped agent is a composable capability. One agent can be joined to another via an *.ampersand.json spec. Code doesn't compose like that — you can't snap two pull requests together and get a bigger feature.

Reputation as a first-class input

Code review checks syntax and tests. Agent review requires trust — how did this actor behave in production? FleetPrompt scores every agent; the deploy gate only opens if trust exceeds threshold. Code factories have no equivalent primitive.

Continual-learning benchmarking

Code passes or fails tests. An agent's quality is a moving target across 9 dimensions: consistency, generalization, forgetting, transfer, plasticity, stability, efficiency, calibration, safety. PRISM watchtowers evaluate each shipped agent continuously. Code factories ship and forget.

Memory that spans deployments

Graphonomous holds the graph of every agent, every outcome, every trust update. When a new agent is built, the factory remembers which patterns succeeded, which failed cyclically (κ > 0), and which deserve promotion.

Spatial governance of a fleet

Reviewing 10 PRs is a list. Governing 103 deployed agents is a city. RuneFort's spatial UI scales where dashboards collapse — districts, forts, rooms, runes. Walls thicken where Delegatic policy tightens. You can see the fleet.

The shift, stated plainly

Spotify Honk is a Dark Code Factory — brilliant, shipping 650 PRs/month, but each PR is terminal. RuneFort is a Dark Agent Factory — each ship adds another autonomous producer to the fleet. The comparison isn't Honk vs RuneFort. It's Honk the line worker vs RuneFort the foreman building more line workers.

Your factory runs in the dark.
RuneFort turns the lights on and runs the conveyor belt.

Autonomous software factories — spec in, tested code out, no human in the loop — are real and shipping production code today. The question is no longer “can agents build software?” but “can you see what they're doing?”

L0
Manual
Human writes everything
L1
Delegation
AI handles discrete tasks
L2
Pairing
AI copilot, human drives
L3
Review
AI writes, human reviews
L4
Spec-Driven
Specs in, code out
L5
Dark Factory
Fully autonomous

The dark factory (today)

  • Specs go in, tested software comes out
  • No human writes or reviews code
  • Holdout validation separates train/test
  • Spotify Honk ships 650+ PRs/month into production
  • StrongDM: 3 engineers, 32K LOC, zero code review
  • YAML DAGs orchestrate single-repo pipelines

RuneFort runs and renders it

  • Dark Factory loop — watches signals, triages, sequences SpecPrompt → Agentelic → FleetPrompt → deploy
  • Spatial observability — walk through the pipeline, not scroll a log
  • PULSE temporal algebra — phases + cadence + typed cross-loop tokens (no equivalent elsewhere)
  • κ-aware routing — cyclic failure detection triggers deliberation
  • PRISM watchtowers — 9-dimension continual-learning evaluation
  • Graphonomous memory — factory learns across sessions, promotes reliable templates
RuneFort Dark Factory Orchestration Loop
retrieve watch_signals GitHub pushes · Graphonomous outcomes · issues
route triage_classify heuristic classification + κ topology check
act pipeline_execute SpecPrompt → Agentelic → FleetPrompt → deploy gate
learn learn_outcomes OutcomeSignal + ReputationUpdate CloudEvents
consolidate consolidate_patterns merge failure patterns · promote templates

Five PULSE phases, running concurrently with runefort.spatial_render. No LLM calls from RuneFort — pure heuristic triage + MCP sequencing. Synthetic mode for demos, R! for real pipelines against live MCP servers.

Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.

B2B privacy boundary: You see full detail inside your own fort. Partner forts appear as opaque shapes — outer walls, gates, and bridge connections, but never their rooms.

From your code to a walkable fort

No new config language. No manual to read. If your system already declares a PULSE loop, it already has a fort — RuneFort just renders it.

  1. 01 ~10s

    Point at a manifest

    Drop in a *.pulse.json or a GitHub URL. Phases become rooms, connections become corridors, invariants become walls.

    pulse://graphonomous.memory_loop
  2. 02 ~20s

    Connect a Graphonomous MCP

    Rooms light up with live node counts, confidence values, and κ topology. Cold rooms dim. Hot rooms glow. No dashboards.

    graphonomous://workspace/default
  3. 03 ~30s

    Press R!

    The dark-factory loop runs: SpecPrompt validates → Agentelic builds → FleetPrompt trusts → deploy gate. Every stage visible as a corridor tile.

    runefort.dark_factory // running

Five levels of resolution

The same tile grid renders differently at each level. Zooming in reveals more; zooming out collapses detail into summary. No special-case rendering — one primitive, infinite depth.

L0 — District View flow
Press enter or space to select a node. You can then use the arrow keys to move the node around. Press delete to remove it and escape to cancel.
Press enter or space to select an edge. You can then press delete to remove it or escape to cancel.
District Manifest json
{
  "district": "runefort_ecosystem",
  "forts": [
    { "id": "graphonomous",  "loop": "graphonomous.memory_loop",      "schema": "(sqlite)" },
    { "id": "webhost",       "loop": "webhost.deploy_invoke",         "schema": "webhost.*" },
    { "id": "bendscript",    "loop": "bendscript.knowledge_edit",     "schema": "kag.*" },
    { "id": "agentromatic",  "loop": "agentromatic.deliberation",     "schema": "orchestrate.*" },
    { "id": "delegatic",     "loop": "delegatic.governance",          "schema": "govern.*" },
    { "id": "agentelic",     "loop": "agentelic.build_pipeline",      "schema": "agentelic.*" },
    { "id": "fleetprompt",   "loop": "fleetprompt.marketplace",       "schema": "fleet.*" },
    { "id": "geofleetic",    "loop": "geofleetic.fleet_learning",     "schema": "geo.*" },
    { "id": "ticktickclock", "loop": "ticktickclock.temporal_learning","schema": "temporal.*" }
  ],
  "shared_ground": {
    "schema": "amp.*",
    "tables": ["profiles", "workspaces", "members", "entitlements"]
  },
  "bridges": [
    { "from": "graphonomous",  "to": "prism",       "tokens": ["OutcomeSignal"] },
    { "from": "prism",         "to": "fleetprompt",  "tokens": ["ReputationUpdate"] },
    { "from": "agentromatic",  "to": "prism",        "tokens": ["OutcomeSignal"] },
    { "from": "deliberatic",   "to": "agentromatic", "tokens": ["DeliberationResult"] },
    { "from": "delegatic",     "to": "*",            "tokens": ["PolicyEnforcement"] },
    { "from": "agentelic",     "to": "fleetprompt",  "tokens": ["AgentPublish"] },
    { "from": "agentelic",     "to": "webhost",      "tokens": ["Deploy"] }
  ],
  "bedrock": "opensentience (OS-001 through OS-010)"
}

One floor plan. It reveals itself.

You don't need to learn eighteen things before you start. The default view is just structure. Everything below is what’s available when you want more — reached by hovering a room or clicking a wall, not by opening a menu.

Structure
Rooms, walls, corridors, gates. Always on.
F
Flow
Animated arrows showing data and token movement.
T
Thermal
Activity heatmap. Hot is busy, cold is dormant.
P
Temporal
PULSE timing. Which rooms are in which phase now.
D
Diagnostic
PRISM scores, anomaly markers, failure clusters.
R
Rune
The dark code. Configs, policy rules, causal chains.
C
Confidence
Node confidence heatmap. Bright high, dim low.
K
Topology
SCC highlighting, κ values, cycle visualization.

Hover a corridor → Flow fades in. Walk near a busy room → Thermal glows. Click a wall → Rune reveals. No overlay menu — motion is the menu.

Forts inside forts

A fort can appear inside any room. That nested fort can contain another. There is no depth limit — the zoom levels repeat recursively. When you zoom deeper, you find another fort. Always.

PRISM fortprism.benchmarkComposeInteract ▽ObserveGraphonomous fortRetrieveRoute ▽ActDeliberationOpenCommit

No depth limit

Zoom L0 → L4 describes one fort. But L4 inside a nested fort becomes L0 of its own district. You never run out of layers to zoom into — you just find more forts.

[]

Declared in PULSE

inner_loops[] is how nesting is made explicit. The trigger condition determines when the nested fort activates. The wait strategy determines whether the outer fort pauses or fires and forgets.

Parent / child, not wrapper / wrapped

A fort inside a room is not embedded UI — it is a real running loop whose lifecycle is governed by its parent. B2B boundary: partner forts render as opaque shapes. Nested forts you own render in full detail.

Every product gets a fort

Every product in the [&] ecosystem maps to a fort. The layout is generated from PULSE loop manifests, [&] capability bindings, and Supabase schemas. If it declares a loop, it gets a fort.

ᚲᚨᚠᛃᛞ
Graphonomous
Memory substrate. Nested in every fort.
Delegatic
Governance. Its walls are every fort's walls.
TickTickClock
Temporal intelligence. The district clocktower.
AgenTroMatic
Multi-agent orchestration. 7-phase deliberation.
Deliberatic
Argumentation protocol. Merkle-chained evidence.
PRISM
Diagnostic watchtower. 9 CL dimensions.
WebHost.Systems
Hosting layer. Deploy and invoke agents.
Agentelic
Engineering. Spec-driven build pipeline.
FleetPrompt
Marketplace. Trust-scored agent registry.
BendScript
Knowledge editor. Canvas-based graph UI.
GeoFleetic
Spatial intelligence. Federated fleet learning.
SpecPrompt
Standards. Git-based spec lifecycle.

Built with what works

SvelteKit + Svelte Flow

Interactive node graphs with pan, zoom, and custom node rendering. SSR-compatible and accessible. Cloudflare Pages for zero-cost hosting.

MCP Data Sources

Live data from Graphonomous MCP and PRISM MCP. PULSE manifests parsed at build time for fort skeletons. RuneFort never makes LLM calls — pure visualization.

Its Own Loop

RuneFort learns from how you navigate. Frequently visited rooms move closer. Unused rooms recede. The fort is itself a continual learning system.

Hidden machinery inside the walls

Runes are the machine language. The blueprint is the human language. The fort should be immediately legible to anyone — “that's a room, that's a hallway, that's a locked door.” Runes reward deeper inspection but never gatekeep understanding.

ᚲᚨᚠ

Fast Path

Retrieve → Route → Act. The standard execution path through a fort. No deliberation needed, knowledge flows directly to action.

ᚲᚨᚱᚠ

Deliberation Path

When κ > 0, cyclic knowledge triggers deliberation. The corridor pulses. Evidence is weighed. Only then: action.

ᚺᛉᛗ

Escalation Path

Disruption detected → diagnostic shielding → human escalation. The tower fires. Someone gets paged.

ᛃᛞ

Feedback Closure

Learn → consolidate. Outcomes harvested, edges strengthened, weak nodes pruned. The fort remembers what worked.

ᚷᚨᚷ

Bridge Relay

Exchange → route → exchange. A CloudEvent token crosses a B2B bridge, gets routed, crosses back. Partner rooms stay opaque.

ᛁᛞ

Memory Revival

Sealed → breakthrough. A cold room thaws when new evidence makes its dormant nodes relevant again. Dagaz breaks the ice.


24 Elder Futhark runes — hover for meanings

Your autonomous workflows
deserve a control room

Whether you're running a dark factory, orchestrating multi-agent workflows, or building the next cognitive architecture — if it has a loop, it gets a fort.

ᚲ ᚨ ᚱ ᚠ ᛃ ᛞ