Three Pillars

Prism has three goals on the developer side: ship faster, use tokens well, and get better at AI coding. The product groups its capabilities into three pillars that serve those goals.

Goal	Score	Pillar that moves it
Ship faster	Speed	Prompt Advisor (saves turns)
Use tokens well	Efficiency	Usage Intelligence (cost visibility) + Prompt Advisor (shorter paths)
Get better at AI coding	Skill	Prompt Advisor (coaching) + Usage Intelligence (habit tracking)

Prompt Advisor

Audience: individual developers.

Helps you write better prompts and improve your AI-collaboration habits in real time and after the fact.

Feature	Description
Context-management nudges	Single-line /compact / /clear suggestions from the submit hook when a session’s turn count or context growth crosses a threshold
On-demand prompt review	/prism:advisor "prompt" returns a PES-scored rewrite with concrete file paths and expected behavior
Session advisor cards	LLM-authored coaching cards per sub-session, tied to the weakest Skill input — see Advisor & Summaries
Session summaries	Short narrative of what happened in a session, surfaced on the explorer and inside reports
Skill profile	Speed / Skill / Efficiency with per-input trends — see PRISM Scores
My Report	Chapter-based personal review with strengths, weaknesses, worst prompts, and heatmaps — see My Report

Usage Intelligence

Audience: developers first; team leads for aggregate views.

Turns raw telemetry into cost and habit signals you can act on.

Feature	Description
Token & cost tracking	Per-session, per-model breakdown with cache-aware cost — /insights/tokens
Throttle detection	Rate-limit events, latency spikes, and recurring drop-off — /insights/throttle
Model rightsizing	Flags sessions where a cheaper model would have sufficed — /insights/rightsizing
Behavioral insight pages	IE, CRR, FC, QR, RLR drill-downs under /insights/*
Session explorer	Browse and drill into any session, turn-by-turn — /my/explorer
Trend analytics	Daily / weekly / monthly views of usage, active time, and Skill inputs

Governance — under development

Audience: team leads and platform engineers.

Caution

Governance is still under development. The pillar is scoped below for context, but the features listed aren’t available yet. Ship dates are not yet committed.

Policy and guardrails for AI-coding tool usage at the team and org level. Governance will run in the Optra gateway and activate only when gateway routing is enabled.

Planned feature	Description
Budget caps	Soft warnings and hard blocks at org / team / developer spend limits
Guardrails	DLP, PII detection, prompt-injection guard, content filtering
Model access control	Restrict which models developers can use
Cost centers & tiers	Attribute spend and assign spending tiers by role or project
Rightsizing recommendations	Data-driven policy suggestions with estimated savings

See Gateway Routing for how gateway mode works today — the routing is in place; the governance layer on top is what’s still being built.

How the pillars work together

Developer writes prompt
    → Prompt Advisor scores it, nudges if needed
    → Request routes through the gateway (if enabled)
    → Telemetry captured (Ingest → Engine)
    → Sub-session completes → SSE + Skill updated
    → Usage Intelligence surfaces patterns + recommendations
    → Developer improves → cycle repeats

Better prompts (Advisor) raise PES, which raises SSE, which lifts Skill and lowers tokens-per-hour (Efficiency). The three pillars aren’t separate products — they’re three views of the same loop. Governance will plug into the gateway hop once its feature set lands.