Tiers and Model Interpolation

The thesis

Same prompts, same pipeline, intelligence scales with the underlying model tier.

Priority prompt compilation makes this mechanical. When the same PromptBlock tree is compiled for a smaller-context model, low-priority blocks get truncated — output degrades gracefully without rewriting prompts. When compiled for a larger-context model, all blocks survive — output is maximum quality.

One prompt, three tiers, same source of truth.

Free tier — the full forge

Everything ships free:

• Full 7-stage standard_build pipeline
• All 8 agent roles (Scout, Knight, Warrior, Prover, Sage, Bard, Wizard, Herald)
• Full dual_scout / triple_scout research workflows
• Full priority truncation + U-shape ordering
• Full gamification layer (XP + respawn + stage tracker + feeding animation)

Model tier: baseline open-weight or cache-optimized entry model.

The free tier is not a demo. It is the product. Upgrade triggers sit above the engine, never inside it.

Pro tier — model interpolation + vault

Same forge, upgraded model tier (Sonnet-class or equivalent).

Individual local vault rolls out after Phase 0 dogfood evaluation — not at launch. It lights up when the founder's personal evaluation confirms the vault earns its keep.

Enterprise tier — collective intelligence

Same forge, Opus-class inference, Opus Wizard review pass, collective shared vault (LanceDB Cloud).

Rolls out after Pro proves stable. SSO and audit logs are Phase 3.

Plugins — the monetization surface

Bonfire monetizes by selling enhancement, never by gating the engine. Plugin categories include:

• Richer TUI experiences
• UI/UX skins and themes
• Theme progression (ember, blaze, whiteheat visual sets)
• Deeper gamification modes
• Integrations and marketplace additions

Plugin architecture is post-v1 work. The architecture spike is filed and scoped.

Honest status