166 lines
7.1 KiB
Markdown
166 lines
7.1 KiB
Markdown
# Epiphany Investor Brief
|
|
|
|
Status: June 2026 discussion packet
|
|
Audience: investors, design partners, and diligence reviewers
|
|
|
|
## One Sentence
|
|
|
|
Epiphany is a project-native agent control plane that makes AI coding work
|
|
inspectable: the agent must show its map, preserve evidence, separate roles,
|
|
survive context loss, and route real changes through reviewable authority
|
|
instead of pretending the chat transcript is a brain.
|
|
|
|
## Why It Matters
|
|
|
|
Most coding-agent products sell speed. Epiphany is aimed at the more expensive
|
|
failure: agents can keep producing plausible local edits after they have lost
|
|
the global architecture. That failure burns review time, creates adapter piles,
|
|
and leaves teams unable to tell whether the next tool swing still belongs to
|
|
the machine being built.
|
|
|
|
Epiphany turns that hidden state into shared objects:
|
|
|
|
- objective and active work
|
|
- architecture and dataflow maps
|
|
- scratch versus durable memory
|
|
- source-grounded evidence
|
|
- role lanes for implementation, modeling, research, verification, planning,
|
|
Persona, and reorientation
|
|
- runtime receipts, tool receipts, and model receipts
|
|
- review gates before findings become project truth
|
|
- compact operator surfaces for humans and future clients
|
|
|
|
The product thesis is not "the model writes more code." The thesis is
|
|
coherence-preserving agent labor: a system where humans can inspect what the
|
|
agent believes, why it believes it, who authorized action, what evidence exists,
|
|
and what should happen next.
|
|
|
|
## What Exists Now
|
|
|
|
Epiphany is already more than a pitch deck. The current repo contains:
|
|
|
|
- typed Rust state and policy in `epiphany-core` and `epiphany-state-model`
|
|
- a Codex compatibility bridge rather than a Codex-owned product brain
|
|
- CultCache `.cc` stores for runtime, heartbeat, agent, local Verse, memory
|
|
graph, and thread state
|
|
- provider-neutral model request/event/receipt documents
|
|
- typed tool capability, invocation intent, and invocation receipt documents
|
|
- runtime-spine job and worker-result documents
|
|
- Mind, Substrate Gate, Eyes, Hands, Soul, Continuity, Persona, and heartbeat
|
|
contract surfaces
|
|
- local operator commands for status, coordinator runs, smoke checks, and
|
|
CultMesh/Verse context
|
|
- a Modeling whitepaper that maps the body, owners, inputs, outputs,
|
|
invariants, bridges, and current cut line
|
|
|
|
The key architectural move is extraction: Epiphany began as a Codex-derived
|
|
harness, but the product architecture is moving into typed Rust, CultCache,
|
|
CultMesh, and CultNet documents. Codex is being narrowed to the honest
|
|
compatibility spine for subscription auth, model transport, and useful host
|
|
affordances.
|
|
|
|
## Where Bifrost Fits
|
|
|
|
The GameCult investment dossier raises the proof bar. Epiphany should not be
|
|
sold as a local demo. The investor-ready loop is Bifrost-first:
|
|
|
|
```text
|
|
Bifrost topic or work item
|
|
-> scoped Epiphany execution request
|
|
-> bounded role work
|
|
-> artifacts and receipts
|
|
-> maintainer review
|
|
-> accepted or rejected outcome
|
|
-> cost, review load, credit, and public-safe proof
|
|
```
|
|
|
|
Bifrost owns public work records, dispatch packets, receipts, credit, reward
|
|
pressure, and governance. Epiphany owns bounded execution, typed memory,
|
|
coordination, verification pressure, and operator-safe proof artifacts. Discord,
|
|
VoidBot, repo Personas, and scripts may mirror or initiate pressure, but they
|
|
must not become shadow governance.
|
|
|
|
## Recent Public Narrative
|
|
|
|
Recent GameCult public material gives investors a clean translation layer:
|
|
|
|
- `The Free Mouth And The Native Body` frames VoidBot as the free live mouth
|
|
and Epiphany as the native high-leverage body.
|
|
- `Epiphany In The Interview Chair` gives the job-interview pitch: Epiphany
|
|
makes an AI coding agent show its map, prove its cuts, survive context loss,
|
|
and stop when it no longer understands the machine.
|
|
- The June 2026 GameCult site project pages frame Epiphany as active harness
|
|
work, EpiphanyAquarium as the operator interface, Bifrost as public work and
|
|
receipts, and Heimdall as identity/grants.
|
|
- The daily damage report turns repo motion into a public work log, which is
|
|
useful raw material for future Bifrost receipts.
|
|
|
|
The investor wrapper should lead with ordinary words: agent work, receipts,
|
|
review, cost accounting, identity, governance, and accepted artifacts. The
|
|
mythic language is internal doctrine and studio voice; the investable object is
|
|
measured human/agent production.
|
|
|
|
## Value Proposition
|
|
|
|
For developer teams:
|
|
|
|
- less review time wasted on incoherent agent output
|
|
- persistent project memory across long tasks and context loss
|
|
- visible role separation between modeling, implementation, research, and
|
|
verification
|
|
- safer authority boundaries for tools, file edits, and commits
|
|
- artifact and receipt trails suitable for audits and postmortems
|
|
|
|
For GameCult:
|
|
|
|
- a native agent substrate that can absorb the live lessons from VoidBot
|
|
- public/project Personas without hidden state authority
|
|
- Bifrost-routed proof work with receipts and credit
|
|
- a commercially licensable control plane while preserving free/reference
|
|
layers where appropriate
|
|
|
|
For investors:
|
|
|
|
- a differentiated wedge in AI-native work governance, not just code
|
|
generation
|
|
- measurable diligence targets: accepted useful work per human review hour,
|
|
cost per accepted artifact, fresh-repo success rate, and sealed public-proof
|
|
export quality
|
|
- optional commercial paths through enterprise Epiphany/Bifrost services,
|
|
source-available licensing, support contracts, or mission-aligned affiliates
|
|
|
|
## Proof To Ask For
|
|
|
|
The next diligence pass should ask for live evidence, not vibes:
|
|
|
|
1. Fresh-repo Epiphany demo with scoped objective, typed state, role routing,
|
|
artifact output, review, and receipt.
|
|
2. Bifrost demo showing topic/work item, dispatch packet, agent transport,
|
|
receipt, ledger or credit record, and outcome.
|
|
3. Cost accounting: model spend, role time, human review load, accepted
|
|
artifact count, rejected-output reasons, and lessons added to memory.
|
|
4. Public/private export proof: no raw worker thoughts, transcripts, private
|
|
notes, secrets, or operator context in public artifacts.
|
|
5. Security model for identity, secrets, write permissions, revocation,
|
|
external repo access, and public publishing.
|
|
6. Design-partner proof on at least one external or semi-external repo without
|
|
supervisor contamination.
|
|
|
|
## Packet Contents
|
|
|
|
- `docs/epiphany_body_whitepaper.pdf`: shareable architecture whitepaper.
|
|
- `docs/epiphany_body_whitepaper.tex`: source for the whitepaper.
|
|
- `notes/epiphany-investor-readiness-roadmap.md`: internal readiness roadmap.
|
|
- `E:\Projects\gamecult-site\docs\gamecult_integrated_dossier.tex`: broader
|
|
GameCult / Epiphany / Bifrost investment dossier.
|
|
- `E:\Projects\gamecult-site\GameCult\Blog\the-free-mouth-and-the-native-body.md`
|
|
and
|
|
`E:\Projects\gamecult-site\GameCult\Blog\epiphany-in-the-interview-chair.md`:
|
|
public narrative source material.
|
|
|
|
## Bottom Line
|
|
|
|
Epiphany is valuable if it can repeatedly turn ambiguous project work into
|
|
bounded agent labor with inspectable state, receipts, review, cost accounting,
|
|
and accepted artifacts. The whitepaper proves the body is becoming legible.
|
|
The investor proof is the Bifrost-first loop producing useful external work.
|