- source inventory and provenance
- immutable corpus versions
- retrieval profile and evaluation
- citations, limits, and rollback
For AI product teams
Ship the AI feature. Stop rebuilding its knowledge layer.
CorpusMesh maintains source-backed knowledge and returns retrievable context. Your models, prompts, agents, and application stay under your control.
Private beta. Access is reviewed manually for a named application. There is no public signup or public retrieval API.
Knowledge maintenance
eu-ai-act-referenceOfficial sourceEUR-Lexmonitored source set
Corpus version2026-08-21.1source verified
Retrieval contractPOST /v1/retrieveprivate beta
Returned to your application
passagescitationsversiondecisionThe recurring job
Your prototype is not the maintenance plan.
Finding a useful passage once is easy to demo. Keeping the source boundary current, reproducible, and safe to serve is the ongoing product work.
- 01
Define the source boundary
Name the authorities, documents, jurisdictions, languages, dates, and exclusions the feature is allowed to use.
source inventory - 02
Preserve proof of origin
Keep the original artifact, reuse evidence, canonical URL, checksum, and transformation history attached to every version.
provenance record - 03
Prepare retrievable knowledge
Normalize structure, add metadata, create deterministic chunks, and build a projection without making the vector store canonical.
immutable version - 04
Measure retrieval before activation
Replay a frozen fixture set, test citations and abstention, then activate only the version that passes its boundary.
evaluation gate - 05
Operate the next source change
Detect upstream changes, rebuild deliberately, preserve the prior version, and keep rollback available.
maintained service
A deliberate boundary
Knowledge on one side. Product intelligence on the other.
CorpusMesh does not take over the part of your application that makes decisions. It makes the knowledge input inspectable and consistent.
Request
CMquery + allowed filtersResponse
context + citations + decision- model and provider
- prompts and agent logic
- customer experience
- the final answer or action
Build versus buy
Build the part that differentiates your product.
CorpusMesh is useful only when the recurring knowledge work costs attention without creating product advantage. There are cases where keeping the whole pipeline internal is the right call.
| Responsibility | Run internally | Use CorpusMesh |
|---|---|---|
| Source intake | Your team finds sources, documents rights, and defines completeness. | A reviewed source inventory defines the bounded corpus. |
| Change handling | Your team monitors every authority and reconciles each material update. | Source checks create a new candidate version only when content changes. |
| Retrieval quality | Your team owns chunking, embeddings, evaluation fixtures, thresholds, and regressions. | A versioned retrieval profile must pass the corpus evaluation before activation. |
| Provenance | Your team designs and preserves the path from each result back to the source. | Source identity, version, metadata, checksums, and citations travel with results. |
| Serving boundary | Your team implements authorization, filters, quotas, metering, and rollback. | The private-beta entitlement pins the permitted corpus, version, filters, and limits. |
Reviewed evidence
Inspect one retrieval all the way back to EUR-Lex.
The EU AI Act Source Trace is an internal engineering reference. It is fixed to a verified corpus version and a reviewed runtime result. It is not customer proof and it does not make the service public.
- Corpus version
- 2026-08-21.1
- Frozen fixtures
- 80
- Boundary violations
- 0
- Official source
- EUR-Lex
- Published evidence status
- Verified internal retrieval
Evaluate the fit
Follow the question to the right contract.
Questions from AI teams
Know the boundary before you integrate.
Does CorpusMesh run our model or generate the final answer?
No. Your team controls the model, prompts, agents, product behavior, and final output. CorpusMesh returns bounded passages, source metadata, citations, the corpus version, and a retrieval decision.
Why not build the knowledge pipeline ourselves?
You should when the corpus is a core proprietary advantage, must remain inside a boundary no provider can enter, or needs highly custom processing. CorpusMesh is intended for teams whose recurring source maintenance is necessary but does not differentiate their product.
Can we send production traffic today?
There is no public signup or generally available retrieval service. The prepared EU AI Act service is a manually reviewed private beta for a named application and entitlement.
Which knowledge is available?
The verified English EU AI Act corpus has a prepared private-beta boundary. Other catalogue entries are research or planning hypotheses and must not be treated as available products.
How can we inspect the source behind a result?
The public EU AI Act Source Trace shows one fixed request, its enforced scope, ranked passages, official EUR-Lex citations, version, evaluation context, and lineage checksums.
Your knowledge problem
Tell us what your application has to keep current.
Name the sources, jurisdictions, update burden, intended workflow, and expected retrieval volume.