AI · the glass box

Wired into every page.Cited every time.

It reads every page, calls every tool, and links every claim back to a source.

Start freeSee an answer, traced →
  1. 8 AI surfaces shipped, each native to the page it lives on.the suite →
  2. Every tool call kept, one click below the answer.one answer, traced →
  3. 4 providers, one interface, routed to the smallest sufficient model.the routing →

One answer, traced

One answer, traced

An illustrative assistant answer in the terminal workspace: one question about NVDA's last earnings, answered in prose with a source chip after every claim. Selecting a citation highlights its source in the ledger beside the answer. Answers are generated at query time in the product, not hard-coded.

workspace / NVDA · assistantroute claude-sonnet-5-5
What changed at NVDA in the last earnings?

Data-center revenue came in +94% YoY and Hopper sell-through is tightening into the back half . Management softened its China guidance while leaving FY27 capex plans unchanged .

On the guide: gross margin holds near 74% through FY27, networking crossed a$10B quarterly run rate , and a new $50B buyback authorization lands as a floor under the print.

Since the print: a four-officer insider cluster, $312M in sells, and net inventory days +6 QoQ.

grounded · every claim carries its source5 sources · 9 citations
Fig. 1 · one answer, traced · illustrative composite; in the product, answers generate at query time from live rows

The suite

Every surface, by the model that runs it.

Each surface is native to the page it lives on, and routed to the smallest sufficient model.

gpt-6-lunaFast classification and single-table lookups. If OpenAI fails, the chat and news enrichment fall back to Claude Haiku 5.5, the rest to Sonnet 5.Smart alertsZero-shot classification
SonnetCross-table synthesis with citations; the Pro default.Morning BriefAsk the marketStock intelligenceLiving Analyst Notes
Opus 5.5Multi-source, thesis-level reasoning.Deep researchbetaAgentic researchin the lab
VoyageEmbeddings, not generation.Semantic searchbeta

Live unless marked.

The difference

A chatbot answers. A glass box shows its work.

QuestionA generic chatbotThe glass box
Where the answer comes fromA generic chatbotModel memory plus a web search. The trail ends at a list of links.The glass box~40 tools over the terminal's own tables. The trail is the answer.
How claims are sourcedA generic chatbotCitations are optional, and a model can invent one as fluently as prose.The glass boxEach answer keeps its tool calls and what each returned, so a claim can be checked.
When the data is missingA generic chatbotThe gap is filled with the most plausible continuation.The glass boxThe answer narrows to what it can prove, and says so.

Pipeline

Ask. Reason. Cite.

Every answer runs the same four phases, and every tool call can be inspected.

  1. 01ParseYour question becomes structured sub-queries.
  2. 02RetrieveTool calls pull the smallest evidence set that answers it.
  3. 03SynthesizeA cited answer with inline source chips. No claim without a row.
  4. 04CacheIdentical follow-ups are served instantly; stale on refresh.

Model routing · cost

The right model for the job.

Most queries go to the smallest sufficient model, with an automatic upgrade when confidence is low.

classificationgpt-6-lunaopenaifallbacksonnet-5anthropic
simple lookupgpt-6-lunaopenaifallbacksonnet-5anthropic
chatsonnet-5-5anthropicfallbackgpt-6-lunaopenai
conversationalsonnet-5anthropicfallbackgpt-6-lunaopenai
synthesisgpt-6-lunaopenaifallbacksonnet-5anthropic
deep researchopus-5-5anthropicgroundingsonar-properplexity
semantic indexvoyage-4-largevoyageembeddings · nightly
TierTop modelMonthly queriesDeep research
Free $0gpt-6-luna100
Pro default $29 / mosonnet-5-51500
Pro Plus $79 / moopus-5-5 150 / mo1,00020 / mo

Four providers, one interface, a cross-vendor fallback on every lane. Every call is recorded in the AI usage ledger; the same tools are public over MCP.

Try these prompts

Start anywhere. The agent finds the rest.

Every example is a real prompt. Numbers, citations and tool calls are generated at query time.

prompt historyuser@convexity
>Compare NVDA / AMD FCF yield over the last 4 quarters
└ → table + 2 chart panels · 8 citations
>Which holdings in my portfolio have insider clusters this week?
└ → 3 hits ranked by conviction · 12 Form-4 citations
>Summarize the 8-K language shifts in DHR's last filing
└ → diff vs prior 8-K, highlighted · 2 citations
>Does the recent congressional NVDA call-options cluster suggest tail risk?
└ → 4 PTRs surfaced, weighted by role · cited
>How does NVDA's Convexity Score compare to its semis peer set?
└ → peer cohort + percentile + sparkline · methodology linked
>What is AVGO's supply-chain exposure?
└ → answer · Go deeper on AVGO ↗ sets up a Deep Research report in Studio

One terminal. One AI backbone. Every answer cited.

Founder pricing locked for life: the free plan needs no card, $29/mo for Pro.

Start freeRead the docs