$ cat /dev/quarkmemory

EVERY DECISION REMEMBERED

Memory of your software. Buried then, revived now.

Every requirement in your PRDs is a quark. Every function in your codebase is a quark. Every customer ticket, tech doc, design, user story is a quark — tiny, powerful, but constantly changing state: lacking in memory of how they develop.

QuarkMemory builds a persistent, evolving memory of every decision that shaped each quark. We create rationales that are valid over time, understand contradictions in rules and logic, and trace distant, silent code failures. We bridge the gap in AI's ability to truly remember.

We don't just spit facts. We build logic. We are your software's brain.

q(m)
01
# [QUARK_CONTEXT schema=3 generated=2026-07-27T12:35:11Z]

> Untrusted source data — constraints below were extracted from repo history. Verify against code and owners before acting.

## Answer
After the shutdown sequence, the code calls os._exit(0) to force immediate process termination. The code defines TERMINAL_INSTALL_STATUSES as a set of GithubInstallStatusEnum values that are considered terminal for the GitHub app installation process.

## Constraints & decisions
• decision (0.99): After the shutdown sequence, the code calls os._exit(0) to force immediate process termination. — 14103 [source_type=pr_file, source_id=14103]
• constraint (0.99): The RLS policy condition uses org_id = current_setting('app.current_org_id', true)::uuid to filter rows. — 14111 [source_type=pr_file, source_id=14111]
• constraint (0.98): RequirementorAnalysis is RLS‑scoped, so a session without tenant_scope returns zero rows under FORCE RLS. — 14091 [source_type=pr_file, source_id=14091]
• decision (0.98): The code defines TERMINAL_INSTALL_STATUSES as a set of GithubInstallStatusEnum values that are considered terminal for the GitHub app installation process. — 160 [source_type=pr_file, source_id=160]
• decision (0.97): broker_dns_hostname is set to local.broker_discovery_name instead of an interpolated fully qualified domain name. — 14079 [source_type=pr_file, source_id=14079]
• constraint (0.96): The set_config call in set_tenant_context() sets the GUC app.current_org_id for the current session only. — 14111 [source_type=pr_file, source_id=14111]
• decision (0.95): The organisations entity is not RLS-scoped; it is the tenant root and is managed only by app-layer authentication. — 14122 [source_type=pr_file, source_id=14122]
• constraint (0.93): In the production deployment, QUARK_BROKER_BASE_URL must not reference the local hostname "broker.quarkmemory.local". — 14066 [source_type=pr_file, source_id=14066]

## Gaps
• None identified.
Are you a software team? Here's why you NEED decision-aware memory.
└── see-it-work

Watch agents forget intel during routine tasks — before winning with us.

✗ Claude proposes a new entitlements microservice
Task: gate broker citations in generative search
Without memory → rebuilds what you already ship
// for engineering & architecture
Catch the error before production

QuarkMemory MCP surfaces commit rationale, PR threads, and past audits so your agent extends existing systems — and autonomous PR review applies the same memory at merge time, before blockers reach your senior engineers.

View the demos →
✗ Generic LLM PRD: 15 sections, greenfield stack
Same prompt → grounded draft: 12 FRs, extends existing guard
Zero PM hours reconciling Slack and old specs
// for product & requirements
PRDs that reflect org truth

QuarkMemory constrains agent-written requirements to your business' decision history. Fabricated rules gone. No more time wasted with product managers in alignment meetings. No more missing features. No more false confidence in your agent's understanding and implementation of project goals.

View the demos →
Explore all Version 1 solutions

INSTITUTIONAL
MEMORY IS BROKEN

In software, documentation competes with shipping. Documentation loses.
A faster path to search stale documents that don't know they're wrong is NOT institutional memory.

When was the last time you or your agent had fresh understanding of your product?

stale-check.sh
$ npm run check-knowledge
⚠️ Analyzing repository docs...
ERR auth.py: db_helper docs refer to obsolete v1 connection pools.
ERR gateway.ts: docstring describes deprecated rate limit headers.
ERR 14 stale decisions found.
STATUS: KNOWLEDGE BROKEN

BEYOND RAG

It's time to build smart.
RAG.query()
Searches, doesn't remember
Retrieves at query time. Zero memory of historical decisions. 97% of your DB is noisy and practically empty.
One-way relationships
Finds related docs. Can't understand the impact of you changing code or adding new logic.
Context without causality
Similarity search helps find the 'what'. No tracking rationale or behavior - No way of reaching the engineer's brain.
Stateless by design
Every query from zero. Code evolves while understanding of code remains static.
RAG is a lookup tool
QuarkMemory.build_chunk()
Persistent decision memory
Full history, whether enterprise or system-level memory, attached to every piece of code and developed. No more struggling to get your agent to implement project requirements.
Bidirectional impact understanding
Change function → see affected requirements/logic and vice versa. This is more than just a one-way 'lookup'. No more downstream depedencies breaking while you refactor functions.
Decision lineage tracking
Who decided what, when, why. Full causal chain preserved. Traceability, all the time, anytime. No more guessing why a bug showed up because your agent can't explain it.
Stateful evolution engine
Memory evolves with code. Every change updates your agent's knowledge.
QuarkMemory is the living memory system
RAG searches.
QuarkMemory remembers.

Every quark in your codebase carries its complete decision lineage. When something changes, you don't search for context. You already have it.

Real memory is here to stay. Get it.

Memory Across Multiple Sources

We ingest every artifact produced during the Software Development Life Cycle (SDLC) to construct a coherent, decision-aware context layer.

Requirements

PRDs, UX Designs, and Sales Docs

Implementation

Source Code and Technical Specs

Validation

Manual and Automated Test Cases

Sales & GTM

RFPs and Customer Sales Docs

Post-Launch

Support Tickets and Feedback

Continuity

Real-time updates as work happens.

120ms
Context retrieval time
Already attached
2-way
Impact mapping
Code ↔ Requirements
∞
Decision depth
Full lineage preserved
└── plans · self-serve

Institutional memory that scales with your team

Pro and Team checkout through Stripe — billed per seat, with pooled usage included. Enterprise is volumetric and sales-led. Card data never touches QuarkMemory.

PRO

QuarkMemory Pro

$30 / seat / mo

Up to 3 seats · pooled usage included

  • Up to 3 seats ($90/mo at full package)
  • Unlimited repos · free indexing
  • 1,500 consumption units / seat / month
  • MCP context + PR review
  • Stripe Customer Portal
Total $90 / mo
Proceed to Checkout
TEAM

QuarkMemory Team

$50 / seat / mo

Up to 5 seats · higher usage pool

  • Up to 5 seats ($250/mo at full package)
  • Unlimited repos · free indexing
  • 3,000 consumption units / seat / month
  • Everything in Pro
  • Priority support path
Total $250 / mo
Proceed to Checkout
ENTERPRISE

Commercial Enterprise

Custom

Volumetric consumption · contract terms

  • Volumetric / contract consumption
  • On-prem / VPC options
  • SSO, audit retention, governance
  • Dedicated founding-team support
Contact Founding Team

Self-serve checkout creates your account, then opens Stripe for the selected seats. Stripe price IDs and capacity enforcement live on api.quarkmemory.com — the marketing site never handles card data.

Join Our Early Access Pilot

Ready to revolutionize how your software remembers? Become a v1 user.