ChromaDB · Deck Outline · Proto

What “Proto” means in our convention

Proto = the faithful-recreation baseline. When we engage a client whose materials already exist (a Figma deck, a Pitch.com URL, a PDF), the first whole-deck variant we author in chroma-decks (or any client-decks repo) is Proto — the same content, same order, same structure, ported into the code substrate without enhancement. Proto’s job is to give us:

1. A working code recreation we can verify against the live artifact (“does this say what the client says?”).
2. A baseline that subsequent variants (Enhanced-v1, Enhanced-v2, Audience-LP-v1, Lightning-5min, etc.) can be honestly compared against — both visually and narratively.
3. A “no-yak-shaving” first pass — placeholders are accepted; what matters is that the spine is in code.

Proto is never the version the founder presents from once we’ve shipped an Enhanced variant. Proto exists to be diverged from. It also exists to be recoverable — if Enhanced-v3 turns out worse than Proto along some axis, we want to be able to checkout the proto commit and study what got lost.

Naming convention (new, locked here, applies to all client-decks repos under dididecks-ai/client-sites/*):

• File: {ClientName}_Deck-Outline__{VariantName}.md in dididecks-ai/context-v/narratives/.
• VariantName patterns: Proto, Enhanced-v1, Enhanced-v2, Lightning-5min, Board-Update-v1, Audience-LP-v1, etc. Hyphen-cased, descriptive, suffix the audience/format/iteration.
• The __ (double underscore) separates DocType from VariantName; the _ (single underscore) separates ClientName from DocType.
• Whole-deck routes follow calmstorm’s precedent: /scroll/{deck-slug}/ (the canonical) plus optional /scroll/{deck-slug}/v2/, /v3/ etc. for variants whose whole composition differs. Proto lives at the canonical /scroll/{deck-slug}/.

This answers the open question in [[Dididecks-AI-Slide-Decks-as-Code]] § Versioning Model — the answer is: whole-deck variants are first-class, named, narrated by a Deck-Outline doc per variant, while individual slide variants live in /drafts/{slug}/{slug}-vN.astro per the calmstorm pattern.

Source artifact (Proto’s faithfulness target)

• PDF: corpus/management-supplied/2026-05-11_Chroma-Series-A_MS-Resort.pdf (13 MB, 14 pages, git-ignored)
• Title: “Context and Memory Infrastructure for AI” (Chroma · Series A · MS Resort)
• Capture date: 2026-05-11
• The CEO presents from this PDF live. Proto must read like a code recreation of this artifact, not an enhanced reinterpretation.

Code anchor

• Repo: lossless-group/chroma-decks (private, branch = development)
• Route: /scroll/pitch/proto/ (Vercel preview). /scroll/pitch/ itself is now the variant chooser listing all variants of the Pitch deck (Proto today; Enhanced-v1, etc. tomorrow).
• Slide components: src/components/slides/proto/01..14-*.astro — one per slide, variant-scoped
• Composition: src/pages/scroll/pitch/proto/index.astro (imports + orders the 13 slide components inside <PageAsDeckWrapper>)
• Pinned commit for Proto: 06d5e59 — also git-tagged chroma-pitch__proto. Roll back to this exact state any time.

The 13 slides

Note: the PDF has 14 pages. Slides 5 (bottleneck framing) and 6 (Chroma solution) were merged in code — they read as one narrative beat (“here’s the gap → here’s our answer”) and benefit from being seen together. Whether to re-split them is a candidate divergence for Enhanced-v1.

01 · Cover

File: src/components/slides/proto/01-cover.astro Composition: centered — Wordmark · Title · “Confidential” · 4-icon row on dotted background.

• Title: “Context and Memory Infrastructure for AI”
• Subtitle: Confidential (italic)
• Footer visual: 4 pixel-art icons (yellow boxes / blue brush / green terminal / red airplane) on a dotted grid — currently SVG approximations in public/brand/icons/; needs replacement with Chroma’s actual icon assets.
• Status: Faithful ✓ · Icon assets = placeholder

02 · Opening Statement

File: src/components/slides/proto/02-opening.astro Composition: centered, single large headline.

• Headline: “Agents will automate knowledge work at every organization on earth”
• Status: Faithful ✓

03 · Traction

File: src/components/slides/proto/03-traction.astro Composition: center-aligned — title · 4 stat cards · “Context Rot” line · customer logo grid · “Cloud platform customers” caption.

• Title: “One of the leading brands in developer AI”
• Stats (4): 27k Github Stars · 14M monthly downloads · 50K cloud teams · $2.4M cloud run rate
• Subhead: “Chroma authored the Context Rot report and popularized context engineering.”
• Customer logos (13): xAI · Mintlify · FTAI Aviation · KREA · Skydance Animation · Weights & Biases · Conduit · Qualcomm · Cofounder · The Prompting Company · Isara · Propel · CBS News
• Caption: Cloud platform customers (italic)
• Status: Faithful ✓ · Logos = text-label placeholders (need real SVGs — crawl-fetch-ingest skill)

04 · Competition

File: src/components/slides/proto/04-competition.astro Composition: top-aligned (the only slide retaining align="start") — title · 2×2 quadrant matrix · “How we win” right rail.

• Title: “Competition”
• Matrix axes: X = Component → Full platform · Y = Closed-source ↔ Open-source
• Dots: Opensearch (top-left) · Chroma (top-right, highlighted orange) · In-house (middle) · Pinecone, Turbopuffer (bottom-left) · Labs (bottom-middle) · Glean (bottom-right)
• “How we win” rail (4):
  • vs Vector DBs: Massive up-lift to build around vector databases to get to platform utility
  • vs Labs: Labs can not offer language-model agnostic solutions, will only push their closed gardens
  • vs In-House: Very few teams will successfully roll their own solution.
  • vs Enterprise search: Enterprise search does not control their own database or models; search does not solve the memory problem.
• Status: Faithful ✓ · Dot positions estimated by eye from the PDF — may need micro-nudges

05 · Bottleneck + Chroma Solution (merged from PDF pages 5 + 6)

File: src/components/slides/proto/05-bottleneck.astro Composition: centered, two-column. Connector line (cobalt SVG arrow) from Memory & Context box → right column.

Left column:

• Title: “The bottleneck is no longer reasoning failures, they are context and memory failures.”
• 3-box stack: Agents (Coding, support, product, operations) ↕ Memory & Context (Learning, Unified real-time context layer · spotlighted in cobalt) ↕ Underlying raw data (docs, code, chat, email, agent traces)

Right column:

• Callout: “The missing piece” (cobalt)
• Title: “Chroma makes the context and memory infrastructure for AI”
• Chroma 5-layer block with side annotations:
  • Governance · auth, audit, policy
  • Intelligence · synthesis, querying · ← Powered by Context-1, our SOTA agentic search model
  • Database · serverless, scalable · ← Powered by ChromaDB, the most popular AI-native database
  • Syncing · slack, github, notion…
  • Evals · tests, drift, versioning · ← Authored Context Rot Report

Status: Faithful to the merged narrative ✓ · The PDF’s “Agents + Humans” and “Raw data” framing bands around the Chroma block were dropped in the merge (the left column’s 3-box stack already names that frame). Re-introducing them is a candidate Enhanced-v1 move.

06 · Difficult Problems (was slide 7 in PDF)

File: src/components/slides/proto/07-difficult.astro (file kept at 07- for now; rendering order set by pitch/index.astro imports) Composition: centered — title · 3-box mini-stack left · two-column bullet list right.

• Title: “Context and memory are difficult problems”
• Mini-stack: Agents ↕ Context & memory (cobalt) ↕ Raw Data
• Two columns:
  • The system problem: Scalable and serverless indexing and search · Intelligent agentic search · Correctable · Legible · Auditable · Compounding
  • The data problem: Permission-aware filtering · Conflicting information · Fragmented information · Continuously changing information
• Status: Faithful ✓

07 · Two Segments

File: src/components/slides/proto/08-segments.astro Composition: centered split layout (Teams ↔ Developers), each side with title · lede · sublede · 3-band mini stack diagram.

• Left — Context for Teams: “Agents that know your business” · Focusing currently on engineering teams with 10-100 engineers · stack: Slack, MCP, Web → Chroma → Raw data (Slack, Notion, Agent traces, Github)
• Right — Context for Developers: “Memory and context infrastructure for your own products.” · stack: Your Agent, your App → Chroma → Your user’s raw data (chat, agent traces, email, documents, crm, code, skills, tools, OLTP, OLAP, S3, filesystems)
• Status: Faithful ✓

08 · xAI Case Study

File: src/components/slides/proto/09-case-study.astro Composition: centered — title · sublede · two-column (3-row stack left, pull-quote + powering list right).

• Title: “Case Study”
• Sublede: “Chroma powers the context layer for xAI’s Grok”
• Stack (3 rows): User (queries to Grok) → Grok (xAI agent · xAI mark) → Chroma (context layer · Chroma mark)
• Pull quote: “Now that we use Chroma for hybrid search, we’ve saturated our existing evals.” — xAI
• Powering list: Grok Collections API · Business Connectors · Consumer Connectors
• Status: Faithful ✓ · xAI logo = text label placeholder

09 · Team

File: src/components/slides/proto/10-team.astro Composition: centered — title · 6×2 card grid (12 members).

• Title: “Team”
• Row 1 (6): Jeff Huber (CEO, YC) · Hammad Bashir (CTO, YC, Snap, UC Berkeley) · Matt Brailey (Head of GTM, VP Weka) · Philip Thomas (Webflow, OpenDNS) · Robert Escriva (Lacework, Dropbox, PhD Cornell) · Tanuj Nayak (Yugabyte, CMU DB)
• Row 2 (6): Gabriel Shahbazian (Superhuman, Amazon) · Sicheng Pan (UC Berkeley, Sky Lab) · Kyle Diaz (Datadog) · Itai Smith (Amazon, UC Berkeley) · TJ Krusinski (Meta) · Kelly Hong (UC Berkeley)
• Status: Faithful ✓ · Avatars = initials-in-circles placeholders (need real headshots — crawl-fetch-ingest skill on LinkedIn or company about-page sources)

10 · Business Model

File: src/components/slides/proto/11-business-model.astro Composition: centered — title · 3 tier cards.

• Title: “Open-core, usage-based, enterprise”
• Open-source: Free · ChromaDB, self-hosted · → unlimited usage, → community support, → top of funnel
• Cloud: Usage-based · fully-managed cloud · → zero ops, → zero friction
• Enterprise: Platform-fee + usage · BYOC, SLAs, Support · → SOC 2, SSO, → BYOC dedicated infra
• Status: Faithful ✓

11 · Roadmap

File: src/components/slides/proto/12-roadmap.astro Composition: centered — title · 3 columns each with sub-title, content, mini bar chart.

• Title: “Roadmap”
• Sync — More connectors: today: Github, S3 → tomorrow: Postgres, Clickhouse, Databricks, Filesystems, Google Drive, Notion, Slack · + Enrichment (Sync-1)
• Database — More scale, more workloads: Single Index size bar chart 250M → 100B
• Agent — Faster, customer fine-tuning: Tok/s bar chart 400 → 10k · + Fine-tuning
• Status: Faithful ✓ · Bar proportions estimated by eye

12 · Ask

File: src/components/slides/proto/13-ask.astro Composition: centered — title (large) · two-column list.

• Title: “Chroma is the context layer for agents, raising $12M”
• Chroma today: → Loved by millions of developers · → Full integrated solution · → Proven infra at-scale · → Open-source
• Milestones by Series B: Research → 3,000 tok/s · Engineering → Massive datasets (100B+) · Marketing → 2x OSS community · GTM → $10M run rate
• Status: Faithful ✓

13 · Closing

File: src/components/slides/proto/14-closing.astro Composition: centered — large headline · sub · ChromaMark · 4-icon row.

• Headline: “Every agent needs context and memory”
• Sub: “Chroma is building context and memory infra for AI”
• Visual: ChromaMark · 4-icon row (same as cover)
• Status: Faithful ✓ · Icons = same placeholders as cover

Asset placeholder status (the honest gap)

These don’t compromise Proto’s role (faithful structural recreation). Enhanced-v1 should land them.

Design choices baked into Proto

• Three modes (light / dark / vibrant) scaffolded but presented in light. Dark and vibrant are technically functional; the founder approves none of them yet for live use. Brand-iteration session is a separate beat.
• /scroll/pitch/ is the canonical Proto route. No /v2, /v3 siblings exist yet; Enhanced-v1 will likely live at /scroll/pitch/enhanced-v1/ or under a new deck slug.
• #1a73e8 cobalt is our best guess at the deck’s spotlight blue. Eyeball it against the PDF; we’ll dial.
• Slide 5 = merged page 5 + page 6 from the PDF. A deliberate code-narrative compression. Re-splitting them is on the table for Enhanced-v1.

What Enhanced-v1 might do (candidate divergences)

This list is forward-looking and non-binding — captured here so reviewers of Proto can flag what they want to see Enhanced address.

• Land the missing assets (icons, logos, headshots) — biggest single visual lift.
• Re-cut slide 5 → re-split into 5a (bottleneck) and 5b (solution) if reviewer feedback says the merged slide is too dense in print.
• Tighten copy on slide 4 — “How we win” right rail is currently 4 short paragraphs; could be 1-line per row with a hover/click for the elaboration.
• Replace stat cards on slide 3 with a richer visual** — sparklines, mini line charts showing growth not just point-in-time.
• Add a “What we’ll do with the $12M” slide between 12 and 13 — the PDF skips straight from roadmap to ask; a use-of-funds slide could strengthen.
• Re-author the closing to feel less like a bookend and more like a CTA. Currently “Every agent needs context and memory” + visual matches the cover; could end on a forward-looking line.
• Variant-aware mode default. Some audiences (LPs in print) want light; some audiences (live demos to engineering teams) want dark or vibrant. Bake the mode hint into the deck slug — /scroll/pitch/, /scroll/pitch-dark/ — or surface a “best viewed in {mode}” hint.

Cross-references

• Parent spec: [[../specs/Dididecks-AI-Slide-Decks-as-Code]] — the “decks as code” thesis, including the Versioning Model open question this doc partly closes
• Sibling spec: [[../specs/Dididecks-AI-Visual-and-Diagram-Component-Library]] — the primitive library that competition matrix, stack diagrams, and bar charts on this deck should eventually consume
• Brand notes: [[../explorations/Chroma-Brand-and-Deck-Notes]]
• Three-modes derivation: chroma-decks/context-v/explorations/Three-Modes-Derivation.md
• Plan: [[../plans/Init-Chroma-Decks-Client-Site]]
• Changelog of Proto’s ship: chroma-decks/changelog/2026-05-11_02.md
• The Enhanced-v1 outline (initially a duplicate of this doc): [[ChromaDB_Deck-Outline__Enhanced-v1]]