What is Moresq Corpus?
Moresq Corpus is a governed repo-memory layer for AI coding agents. It helps them retrieve verified implementation patterns, proof receipts and source metadata before regenerating the same work.
Retrieval-first repo memory
Stop agents from solving the same code twice.
Moresq Corpus helps AI coding agents retrieve proven code patterns before spending tokens and engineering time regenerating the same work.
5c
default retrieval fee
20c
trial credit cap
60%
contributor share
moresq-corpus / retrieval sessionproof attached
$ corpus.search "stripe webhook idempotency"
3 verified matches
snippet: stripe-webhook
source: apps/data + packages/corpus
labels: production, billing, tested
$ corpus.retrieve stripe-webhook --budget 700
receipt: cr_27f9
license: internal approved
tokens_saved: estimated 612
miss_backlog: unchanged
Live surface
MCP search
agent-facing snippets with short context budgets
Proof receipts
source, quality, license and apply metadata
Console
wallet, slash commands, misses and retrieval audit
Corpus ledger
deposits, reports, payouts and abuse review
Product facts
Everything relevant in one place.
Moresq Corpus is built for the practical moment when an agent needs to decide: retrieve a known implementation, generate fresh code, or escalate to a human.
What it is
A governed repo-memory layer for AI coding agents, built around reusable cards instead of raw context dumps.
Who uses it
Developers, platform teams, AI coding users, contributors, reviewers and internal Moresq IA/eval pipelines.
What it stores
Implementation cards, snippets, runbooks, source metadata, proof receipts, search misses, corrections and incidents.
How agents connect
MCP search/retrieve/apply flows for Claude Code, Codex, Cursor, Windsurf, Cline and CI agents.
How trust works
Every retrieval carries provenance, license status, quality labels, wallet billing and a revocation/report path.
How it learns
Misses, successful applies, failed retrievals, corrections and rejected deposits become governed learning events.
SEO intent map
Five search territories Corpus should own.
The long-term position is not “AI tool” in general. It is the operational layer where coding agents search, retrieve, prove and reuse implementation patterns before generating from scratch.
Agent repo memory
Memory for AI coding agents across repos and sessions
Own the search intent around Claude Code memory, Codex repo memory, Cursor workflows and reusable implementation decisions.
Verified code snippets
Source-backed snippets with receipts and revocation
Position Corpus against hallucinated boilerplate with provenance, review status, rights and proof receipts.
Retrieval-first coding
Search known patterns before generating new code
Frame the category around lower drift, smaller review surfaces and learning from search misses.
MCP corpus
A governed corpus layer for MCP coding tools
Capture developers comparing internal MCP servers, snippet registries, vector search and productized agent memory.
Private coding knowledge base
Keep team-approved code patterns private and searchable
Explain the private layer with sanitization, review gates, provenance and controlled retrieval for coding teams.
Account and credits
Sign in, claim trial credits, then retrieve verified cards.
The console handles sign in, account creation, wallet sync, launch trial credits, Stripe checkout, retrieval history and contributor payout status from one place.
Sign in
Restore a wallet session, retrieve paid cards, inspect access history and continue after checkout.
Create account
Create a Corpus wallet, claim launch trial credits and start testing search/retrieve flows.
Claim trial
Use the `/trial` command in the console to get trial credits when available.
Buy credits
Open checkout from the wallet panel for starter, pro or scale credit packs.
Brand position
Reusable code patterns, not another generic knowledge base.
Corpus owns the point where an AI agent decides whether to invent, retrieve or escalate. The brand is precise, inspectable and operational: proof before prose, snippets before speculation, missed searches before vague roadmaps.
01
Search the workbench
Agents query repo patterns, runbooks, snippets and prior fixes before drafting new code.
02
Retrieve the receipt
Corpus returns compact context with source, license, confidence labels and billing proof.
03
Apply the known pattern
The agent adapts the few project-specific variables while the product preserves provenance.
04
Capture misses
Every miss becomes a backlog item for better snippets, evals and higher retrieval hit-rate.
Proof receipt
The asset is not the snippet. It is the proof around the snippet.
A retrieved card should show why an agent can trust it, what it cost, where it came from, and what to do if it later fails. That is the difference between memory and governed reuse.
See receipt model ->corpus.receipt cr_27f9
Source
repo path, content hash, version and origin metadata
Quality
trust tier, scan result, usage count and incident state
Rights
license, provenance attestation and revocation status
Economics
credits charged, token estimate, time estimate and receipt id
Operating system
Every repeated implementation becomes a retrievable asset.
Economic thesis
Moresq sells less code. It sells less agentic waste.
The bet is simple: a verified retrieval plus adaptation should cost less than fresh generation, human verification and correction. It is strongest for repeated integrations, boilerplate, tests, scripts, CI, infra patterns and shared team conventions.
Metadata-first retrieval
Search short implementation cards first, retrieve the full code only when the match is worth the context budget.
Developer time saved
Reduce repeated boilerplate, subtle hallucination reviews, CI template rewrites and pattern archaeology.
Retrieve vs generate
Make the agent compare lookup cost and integration risk against fresh generation, validation and correction loops.
Marketplace discipline
Reusable know-how can become an asset, but only when quality, provenance and review status are visible.
Economy signal
The saving is the whole agent loop.
Corpus is not only a token cache. It compresses the full waste loop: repeated prompts, context loading, generation, human review, correction, reruns and the energy behind those extra cycles. Tokens are the visible meter; time and verification effort are the product value.
95h
tracked engineering window
193
reuse moments observed
~570k
central estimate tokens avoided
0.098-4.78 kWh
estimated energy avoided
Estimated internal sample
Ledger sample: 193 reuse observations across roughly 95 hours, 203 agent turns, 48 project surfaces and 21.84M measured local Codex tokens. Estimated corpus savings center around 570k tokens, with a directional range of 430k-711k. The claim is cumulative optimization of the whole agent workflow, not only lower token spend.
Energy
97.9-4,781.9 Wh avoided, with 44.4-2,559.6 Wh as a facility-load proxy.
CO2e
About 11-640 g in an EU-like scenario, up to roughly 1.39 kg in a hyperscale stress scenario.
Water proxy
About 0.004-5.12 L depending on WUE assumptions.
Limits
Cost is not claimed without exact model pricing and input/output split; no physical datacenter build avoidance is claimed.
AI learning loop
Agents learn from receipts, corrections and misses.
Corpus does not silently train on raw chat. It records governed learning events: what was searched, what was applied, what failed, what humans corrected and which snippets were rejected or revoked.
Miss
A failed search becomes a candidate implementation card, not a dead end.
Apply
Successful retrievals strengthen similar future rankings only when tied to receipts.
Correct
Human or agent corrections become repair pairs for evals and future versions.
Review
Incidents and rejected deposits become negative examples before they affect production.
Memory is not enough
Moresq should be compared against waste, not only against memory tools.
Generic agent memory
Remembers facts and conversations
Moresq remembers verified implementation patterns with receipts.
Vector DB / RAG
Retrieves chunks from private docs
Moresq retrieves reviewed cards with license, quality and apply metadata.
IDE agent context
Understands the current repo window
Moresq compounds reusable patterns across sessions, repos and agents.
Private MCP corpus
Maximum control, high setup burden
Moresq adds review queues, proof receipts, learning exports and reuse accounting.
Field notes
Content for agents, terminals and MCP retrieval.
FAQ
The questions Google and users usually ask first.
This section keeps the public explanation short, explicit and aligned with the product surface above.
What does it store?
Implementation cards, snippets, runbooks, source metadata, proof receipts, search misses, corrections and incidents.
How do agents use it?
Agents search first, retrieve a compact verified match, then adapt the few project-specific variables instead of inventing the whole block again.
Why are proof receipts important?
Receipts make retrieval auditable. They show source, rights, quality, economics and revocation status so reuse is governed instead of opaque.
Who is it for?
Terminal coding agents and the developers who use them, including Claude Code, Codex, Cursor, Windsurf and MCP clients.