Docs · Hero Agent
HERO sequence — the seven slots
You pick a specialist; the server resolves that slot to the best currently-available engine. Vendor model ids never appear in the UI — this page states tiers in vendor-neutral terms.
| Slot | Purpose | Provider tier | Context | Cost | Data sent | Eval | Free plan |
|---|---|---|---|---|---|---|---|
| HERO NOVA Everyday answers, fast · nova | chat — general chat, explanations, quick help (the default slot) | Frontier-class hosted (120B preferred) | 131K | $0.001 / call | Prompt + selected context for this call | Manual QA (Phase 0) · eval harness planned Phase M | included |
| HERO FORGE Builds real software · forge | build — code generation and multi-file project builds | Frontier-class hosted (120B preferred) | 131K | $0.005 / call | Prompt + selected context for this call | Manual QA (Phase 0) · eval harness planned Phase M | included |
| HERO SAGE Thinks before answering · sage | reasoning — architecture, hard bugs, trade-off analysis, plans | Reasoning-class hosted (27B preferred · 120B fallback) | 131K | $0.02 / call | Prompt + selected context for this call | Manual QA (Phase 0) · eval harness planned Phase M | metered |
| HERO LENS Finds what's wrong · lens | review — correctness, security, performance and style findings | Frontier-class hosted (120B preferred) | 131K | $0.02 / call | Prompt + selected files for review | Manual QA (Phase 0) · eval harness planned Phase M | metered |
| HERO SCRIBE Writes the words · scribe | docs — documentation, READMEs, comments, commit messages, naming | Fast-class hosted (20B preferred · 120B fallback) | 131K | $0.005 / call | Prompt + selected context for this call | Manual QA (Phase 0) · eval harness planned Phase M | included |
| HERO ATLAS Remembers everything · atlas | long — whole repositories, long histories, giant files | Frontier-class hosted (120B preferred · 27B fallback) | 131K | $0.02 / call | Prompt + large context slice (up to the window) for this call | Manual QA (Phase 0) · eval harness planned Phase M | metered |
| HERO SWIFT Instant and lean · swift | fast — short, snappy answers, chips and completions | Fast-class hosted (20B preferred · 120B fallback) | 131K | $0.001 / call | Prompt + selected context for this call | Manual QA (Phase 0) · eval harness planned Phase M | included |
How to read this. "Provider tier" reflects the engine preference each slot
resolves to today (server-side, from the running sequence definition as of 2026-11-05);
if no configured engine answers, every slot still falls back through the free path — it never
silently swaps vendors in the UI. "Data sent" is per call: only the prompt and the context
attached for that slot travel — see the privacy notice.
"Eval" is honest: automated evaluation is a later phase (Phase M), not something shipped yet.
Scaffold. This page is the Phase-0 slot reference: identities, purpose,
context, cost and data are current; engine-level detail (exact routing, fallback order,
benchmark results) lands in a later phase.