GoodMem

PAIR Systems

Open, self-hostable memory and retrieval
control plane for enterprise AI systems

Enterprises are moving from RAG experiments to agentic systems.
They need a governed memory layer they can actually control.

Lead with the category, not with a feature list. The goal is to define GoodMem as infrastructure for enterprise AI systems rather than as another RAG app or vector database.

Why Now

A market inflection is underway

• The market has shifted from chat demos to agentic systems with persistent context.
• Retrieval quality increasingly determines agent reliability and enterprise trust.
• Enterprises now care about governance, compliance, and deterministic operations around memory.
• Buyers do not want their memory layer trapped inside one cloud or buried inside application code.

What changed

• 2023 made RAG mainstream.
• 2024 and 2025 reframed RAG as long-term agentic memory.
• 2026 budget cycles are now funding infrastructure for enterprise AI control, not just experiments.
• Agent frameworks such as Google ADK make agent building faster and increase demand for a standard memory layer underneath.
• No standard memory layer has won the market yet.

This slide should answer why this company exists now rather than five years ago. The strongest line from the older deck is that the market was not ready in 2020 to 2022, but it is ready now because the agentic stack created a real need for persistent, selective context management.

The Problem

Enterprises are stitching together mission-critical memory by hand

What breaks

• LLMs do not have persistent, controllable memory
• Retrieval quality is inconsistent and hard to improve
• App-by-app memory logic creates operational sprawl
• Security, auditability, and lifecycle controls are often afterthoughts

What enterprise buyers actually need

• Persistent long-term memory
• Deterministic, high-recall retrieval
• Governed and auditable infrastructure
• Provider-neutral deployment and inference control
• Typed APIs that platform teams can standardize on

Keep this slide buyer-centric. The point is not “our architecture is elegant”; the point is that enterprise AI teams are missing a standard control plane for memory and retrieval.

What GoodMem Is

GoodMem is not just a vector database and not just a memory SDK

Memory Layer

• Managed memories
• Retrieval pipelines
• Persistent storage
• Long-term context

Resource Layer

• Managed embedders
• Managed rerankers
• Managed LLMs
• Provider proxying

Control Plane

• Diagnostics and health
• Lifecycle controls
• Extension points
• API-first and self-hostable

GoodMem should be framed as a memory + retrieval + inference control plane for enterprise AI systems.

Built to plug into agent frameworks and provider-neutral enterprise stacks, including Google ADK-based workflows.

This is the core category-definition slide. Everything else in the deck gets easier if this framing is accepted.

Why We Win

More Open

Against managed RAG suites, GoodMem is more self-hostable, less opinionated, and more provider-neutral.

More Complete

Against memory-only products, GoodMem manages not only memory but also retrieval and inference resources.

More Portable

Against hyperscaler-native stacks, GoodMem gives customers a portable control plane across clouds and deployment models.

Competitive shorthand

• Closest direct competitors: Vectara, Contextual AI, Credal, Ragie, Onyx, RAGFlow
• Closest memory-layer competitors: Mem0, Zep, Cognee
• Common alternatives: AWS, Google Cloud, Azure, Glean, Elastic, Coveo

Do not turn this into a feature checklist. Keep it as a positioning slide: open vs suites, broader than memory-only, more portable than hyperscalers.

Retrieval Quality Is the Wedge

The optimization layer is what should make GoodMem hard to replace

What this slide needs to prove

• Domain-specific uplift over a credible baseline
• Why auto-tuning matters commercially, not just academically
• Why customers should trust Pair to improve retrieval without building an in-house IR team

Current narrative

• Off-the-shelf embeddings and generic rerankers do not deliver mission-critical enterprise quality
• Pair’s optimization work should turn GoodMem from infrastructure into an intelligence layer

TODO: replace this box with the latest validated auto-tuner vs Crumb result

• Metric: [fill in]
• Baselines: leading off-the-shelf models (e.g. OpenAI, Crumb) once locked
• Current model / run: [fill in]
• Dataset: [fill in]
• Last updated: [fill in]
• One-sentence takeaway: [fill in]

Do not present this slide externally until the chart is replaced with current numbers. This is intentionally scaffolded as a reminder to insert validated benchmark data.

Early Commercial Proof

$60K

Live recurring ARR today

$460K

ARR if current Incorta infrastructure partnership closes

$90K

Total historical revenue

Evidence that the motion is real

• A Fortune 50 customer is live and paying through the Incorta OEM channel
• Incorta has an unsigned $400K/year infrastructure partnership proposal in motion
• Wanclouds validated paid demand before the category had fully formed
• Agiletek, Slalom, and Nisum support the emerging SI channel

Stay precise here. The strongest credible framing is not “we already have massive ARR”; it is “the OEM motion is live, revenue-bearing, and now expanding into infrastructure.”

Business Model and GTM

Four motions, one platform

• Embedded infrastructure partnerships
  • $300K-$400K+ predictable annual license
• OEM per-instance licensing
  • $25K-$110K annual subscription per end-customer instance
• SI channel distribution
  • enterprise and federal access
• Direct enterprise over time
  • higher ACV once platform trust is established

How enterprise buyers actually adopt

• Customers want infrastructure plus engineering help getting it into production
• Forward-deployed engineers accelerate deployment and partner enablement
• Federal and SI motions may require selective cleared FDE hires
• The model is not to build a large services bench ahead of demand

This slide should make two things legible: how the company monetizes and how customers actually adopt. The new point to emphasize is that buyers are not purchasing AI infrastructure in a vacuum; they need help integrating it into real enterprise stacks.

Market

Control-plane software market, not “all AI spend”

$4.5B-$7.5B

Core TAM

$1B-$2.5B

Near-term SAM

$30M-$100M

3-5 year SOM

• GoodMem’s category is memory, retrieval, reranking, and inference orchestration
• The market model deliberately excludes most end-user assistant spend, hyperscaler infra spend, and services
• The clean midpoint is a roughly $6B software market today

Keep this to one slide. The category definition matters as much as the headline number.

Why This Team

Amin Ahmad, Founder and CEO

• Pioneered neural information retrieval at Google Research in 2017
• Applied retrieval at scale across roughly 15+ Google product efforts
• Founder and original architect of Vectara
• Led product, engineering, and ML from zero to enterprise launch
• Personally closed design partners, early customers, and major enterprise deals

Forrest Sheng Bao, Co-founder and Head of ML

• 10+ years as a published ML researcher
• Former professor and later ML leader at Vectara
• Adds evaluation, optimization, and applied model depth

Selected Team

• Tom Diffenbach: ex-Google L6, senior systems
• Zaid Abdurehman: ex-Oracle, security and hardening
• Rogger Luo and Weisi Fan: PhD-level ML and optimization depth

Make this a founder-market-fit slide, not a roster slide. Amin carries the center of gravity. Forrest is the ML complement. The rest of the team is there to prove execution depth.

The Raise

Raising $10M for 30-36 months of runway to make GoodMem the default memory layer for enterprise agentic AI

Use of funds

• Harden the enterprise control plane
• Push retrieval quality and auto-tuning into a repeatable advantage
• Build deployment acceleration and GTM across OEMs, SIs, and direct enterprise
• Stay capital-efficient: no large sales org, no broad international expansion, no large services bench ahead of demand

24-month targets

• Reach $2.5M-$3.5M ARR
• Get Incorta to >$500K ARR
• Add 1+ additional OEM / platform partner
• Add 5-6 direct customers
• Reach 10+ production deployments

Product proof target

Show measurable retrieval-quality gains on customer-specific data versus leading off-the-shelf baselines once the benchmark artifact is locked.

This slide should now answer what the money buys much more concretely: a longer runway, enterprise hardening, retrieval-quality advantage, and a deployment motion that can turn OEM proof into a diversified business.

The Vision

Install GoodMem wherever enterprises run agentic AI

If the market standardizes on a portable, governed memory and retrieval control plane,
GoodMem can become foundational infrastructure for the next generation of enterprise software.

End on ambition, but keep it tethered to the infrastructure thesis. The point is not “all AI”, it is “the standard memory and retrieval layer for enterprise AI systems.”