Core vs Forged

Forven ships in two tiers. Core is the disciplined local lab: build, backtest, run the gauntlet, and paper-trade strategies on your own machine, with the autonomous research daemon doing the grinding. Forged adds the AI agent layer on top — the brain, the specialized agents, deep-dive chat — running on your own model key.

During the beta, this distinction is mostly academic: beta access unlocks the full product surface. This page explains what each tier includes so the boundary is clear when it starts to matter, and states plainly how Forven treats your model spend.

The short version

The dividing line is the AI layer. Everything that makes a strategy survive a gauntlet on its own merits lives in Core. Everything that puts an LLM in the loop — autonomous ideation, agent tasks, interactive deep-dive refinement — is Forged.

Core — the local lab

Core is the whole research discipline, offline-capable and key-free. You get the complete pipeline from idea to paper trading without ever talking to a model provider:

• Backtesting and the lab. Pick a built-in strategy, set symbol, timeframe, and dates, and run a backtest with a 70/30 in-sample / out-of-sample split. Optimize parameters under constraints.
• The gauntlet. Put a strategy through the robustness battery — walk-forward analysis, parameter jitter, cost stress, regime split — and read its robustness score and verdict before anything touches capital.
• The research daemon. The autonomous background loop that invents hypotheses, spawns child strategies, runs them through the gauntlet, retires the losers, and surfaces a shortlist. It runs on deterministic policy and scheduled jobs — the daemon itself is part of Core.
• Paper trading and the scanner. Simulate execution against the live feed with no capital, with the live multi-strategy scanner generating signals under regime filtering and risk gates.
• Risk controls. Max position size, concurrent-position limits, daily-loss halt, the drawdown kill-switch — the hard controls are always on, in every tier.

In other words: Core is enough to do real, disciplined quant research. The AI layer is an accelerant, not a prerequisite.

A note on the brain and the daemon. The autonomous research daemon loop is Core. The brain — the hub-and-spoke orchestrator that dispatches work to LLM-driven agents — is the Forged layer. The daemon's deterministic scheduling runs without a model key; the brain's reasoning does not.

Forged — the AI agent layer

Forged adds the parts of Forven that think with a language model. It is built on a strict separation of duties:

• The brain — the sole hub-and-spoke orchestrator. It receives all agent output, routes lifecycle decisions, enforces quality gates, and keeps persistent working notes and self-judged lessons. Agents never task each other; everything returns to the brain.
• The agents — four specialized roles: strategy-developer (writes strategy code), simulation-agent (runs backtests and optimizations), risk-manager (gates paper promotion), and execution-trader (governs live signals). A hard guardrail applies across all of them: agents never place trades directly. The execution-trader emits signals; the risk-manager and your approval gate the fills.
• Deep-dive strategy chat — an interactive session for one strategy where you (or an agent) iteratively read code, edit params, and re-run backtests, up to eight tool rounds per turn.
• Bring-your-own-key model routing — Forged runs LLM calls through a configurable provider chain (BYOK) with fallback. You bring the keys; Forven brings the orchestration.

These are the capital-adjacent, expensive parts of the product, which is why they sit behind the tier boundary and behind cost controls.

BYOK — and what Forven never does

The AI layer is bring-your-own-key. You supply credentials for your model providers; Forven routes calls through them. There is one promise worth stating without hedging:

Forven never resells model tokens, and never sees your key. LLM spend is between you and your provider. Forven's job is to route the call and bound the cost — not to mark it up.

Keys live in the local keystore on your machine. The model-routing layer reads them locally to make calls; it does not ship them anywhere. Spend is estimated locally from token counts and bounded by cost controls — an opt-in daily agent cap and a per-thread deep-dive cap. For the full picture of providers, fallback chains, and where keys are stored, see models & providers (BYOK).

Beta — everything is unlocked

For the duration of the beta, the tier boundary is not enforced. The Forven entitlement API reports your tier as forged with a beta status, which unlocks the full product surface — Core and the Forged AI layer:

{
  "tier": "forged",
  "status": "beta",
  "currentPeriodEnd": null,
  "offlineGraceHours": 72
}

That tier: "forged" value is deliberate: current desktop builds gate features on the entitlement tier, and reporting forged during beta means nothing is held back from beta users. The status: "beta" field marks that the account is on beta access rather than a paid subscription. The mechanics of that handshake — the device-code flow, refresh tokens, offline grace — are covered in beta access & device handshake.

So, practically:

• During beta: you have everything. Core and Forged, no extra step.
• You still need your own model key for the Forged layer to do anything. Beta unlocks the features; it does not supply inference. Bring a key, or run Core-only.

What this means for you

• If you want to do quant research without an LLM key at all, run Core. The lab, the gauntlet, the daemon, paper trading, and the risk controls are all there.
• If you want autonomous ideation, agent-driven development, or deep-dive chat, that is the Forged layer — bring a model key, set a cost cap, and let the agents work.
• Either way, the safety floor is identical. Paper-before-live is mandatory, beta builds hard-lock to paper, and the kill-switch is always armed — independent of tier.

Caveats (beta)

• The tier split is not enforced during beta — entitlement reports forged, so both tiers are fully available. Plan around the boundary, but don't be surprised that everything works today.
• Forged features are inert without a model key. Unlocking the AI layer is not the same as paying for inference; that is always your own provider spend.
• Cost figures Forven shows are local estimates from token counts, not your provider's invoice. Reconcile against your provider dashboard.

Forven is a research tool. Results are illustrative, never predictive, and nothing here is financial advice. Neither tier changes that: Core and Forged are about how much help you get building strategies, not about any expected return.

Related

• Agents — the AI agent layer that defines the Forged tier
• Models & providers (BYOK) — bring-your-own-key routing and where your key lives
• Beta access & device handshake — the entitlement flow that reports your tier
• Cost controls — bounding the spend the Forged layer can incur