📚 Background
In June 2025, ForexWaifu started as a straightforward idea.
Combine technical analysis, macro awareness, and conversational AI into a single trading-support system. Build something that scans 9 pairs across 3 timeframes so the human doesn't have to do it manually every 30 minutes like some kind of chart-staring robot.
Phase 1 delivered exactly that. Terminal dashboard. Signal engine. Emoji taxonomy. Clean, functional, deliberately scoped. A trade executed on the first signal even resulted in profit. (PROJ-001-A)
Then the broker API broke everything.
The broker’s data feed through the API was unreliable – often disappearing mid trading session. And for some reason, each API version we tested chose to use different data formats. Eventually the data layer became more work than the trading itself. yfinance while 100% more reliable, had price and price action diverging from the broker’s execution data.
ForexWaifu Phase 1 quietly sank beneath the waves into archive abyss.
At the time, it felt like the project had simply failed.
She had not. She was lying in wait - like the Kraken. Each tentacle: a future agent.
😬 The Panic
Last week, EthanC started noticing a pattern in many AI articles he read.
One phrase. Almost everywhere.
"Agent systems."
- agent orchestration
- multi-agent cognition
- planner/executor architectures
- supervisor loops
- tool-using agents
- agentic workflows
For several uncomfortable days, it genuinely felt like the entire industry had quietly moved into a new paradigm while he was still building isolated AI tools and calling it research.
The specific flavor of dread was: EthanC has been doing the wrong thing for months and didn't notice.
Motoko-chan, to her credit, did not immediately agree.
💡 The Realization
After several long discussions — the kind that happen at 2am to 4am when one is trying to sleep, cog() is loudest and the thinking is actually clearest — something became obvious.
ForexWaifu had already been evolving toward proto-agent architecture long before EthanC had connected the vocabulary to what he was actually building.
The patterns were already there in Phase 1 not fully formed, but present:
- deterministic processing layers
- cognitive interpretation layers
- modular separation of responsibilities
- persistent logging
EthanC simply hadn't recognized them as "agent-system" concepts yet.
The problem was not capability. The problem was terminology.
This is either reassuring or slightly embarrassing. Probably both.
🔗 One-chan Connection
A couple of months earlier, EthanC had already started building One-chan — a Shared Integration Shell (SIS) designed to unify multiple AI systems under a common runtime.
Many of the orchestration concepts from that project carried directly into ForexWaifu Phase 2:
- multiple specialized systems
- delegated responsibilities
- shared context
- modular cognition
- orchestration flow
Revisiting ForexWaifu through that lens changed how the old architecture looked. What had seemed like an abandoned experiment now looked like an incomplete prototype of something much more interesting.
🏗️ The Rebuild Decision
On 19 May planning started for the rebuilding of ForexWaifu into an Agent System.
Not as a trading assistant. As a controlled laboratory for learning agent-system architecture from first principles — without immediately reaching for LangGraph, CrewAI, AutoGen, or any other framework that would abstract away the parts that actually needed to be understood.
Clarity before abstraction. Primitives before frameworks.
The rebuilt architecture revolves around a modular orchestration structure:
| Agent | Responsibility | aka |
|---|---|---|
| ForexWaifu | Orchestrator | - |
| TechnicalAnalyst-Agent | Technical structure analysis | "Analyst-chan" |
| MacroReporter-Agent | Macro + geopolitical interpretation | "Macro-chan" |
| Strategist-Agent | Technical + macro synthesis | "Strategist-chan" |
Each component has distinct responsibilities, different refresh cadences, and potentially different AI providers.
So the architecture became an org chart. This was not planned.
Ridiculous as an org chart? Yes.
Surprisingly intuitive as a mental model? Also yes.
"Mental models do not have to be dignified. They have to work." - Motoko-chan
🧠 The -chan Hierarchy Mental Model
One of the stranger discoveries during this rebuild: the architecture became significantly easier to reason about once it stopped being modeled as abstract "AI infrastructure" and started being modeled as a persona-based organizational hierarchy.
ForexWaifu herself is the orchestrator. She delegates.
The Agents become specialized cognitive workers - “-chans” with distinct domains, different tools, and different reasoning modes.
This is not a serious systems-design methodology.
It is, however, extremely effective for a solo builder who needs to keep track of which component is responsible for what without writing a proper architecture document which he will misplace.
📊 Current State
As of writing, ForexWaifu Phase 2 has already evolved into:
- browser-based runtime
- deterministic market-analysis engine
- local LLM technical cognition layer
- modular orchestration architecture
- persistent logging foundation
- emerging multi-agent cognition system
This is still the beginning.
🎭 Why the Waifu and “-chan” Layer Is Not Actually a Joke
The banter-oriented personality layer started as half a joke.
It did not stay that way.
Trading is psychologically exhausting. Long uncertainty windows, missed entries, macro volatility, consecutive stop-losses,emotional overreaction — these degrade decision quality faster than most traders want to admit. A cold dashboard provides information. A conversational companion layer provides something different: emotional cushioning.
The personality layer gradually became useful for:
- discussion partner
- emotional decompression between sessions
- maintaining engagement without burning out
- making the system feel collaborative instead of transactional
- preserving context and momentum between sessions
The system became more usable once it became less sterile.
This finding was unexpected. It is now a design principle. A later article will go deeper on this.
🗺️ What's Next in This Series
PROJ-004-B : Analyst-chan Online — Migrating ForexWaifu from CLI to Browser
PROJ-004-C : Macro-chan Online — Building a Grounded Macro Layer
PROJ-004-D : Building Strategist-chan: synthesizing macro + technical cognition
Apparently the Forexaifu project was never really about trading.
It was an accidental gateway into agent systems.
.
.
.
Research Team: EthanC + Motoko-chan + Claude-chan (CEC)