Ingest and database
- Multiple accounts
- Project communication
- Attachments
- Thread rebuild
- Contacts and domains
Project material is first stored locally and legibly. No presentation layer and no illusion.
Technology
Pelenoir is local project memory for AEC organisations.
Pelenoir first turns project material into a local, traceable structure.
It collects project communication, attachments, documents, threads, contacts, revisions, commitments, decisions, projects and time.
Then it cleans, segments, labels and connects them into a database, graphs, indices, metrics and an audit trail.
The language model comes only at the end: as an interpreter over prepared and traceable context.
The output is therefore not model opinion. It is an answer that can show what it was built from.
Point of the system
Pelenoir saves time by stopping the company from reconstructing its own past.
A construction project rarely breaks because of one decision. It breaks down through small traces: an unanswered question, postponed approval, unclear commitment, supplier pressure or a person who knows but has no time.
Common tools show what someone disciplined enough entered into them. The most expensive context is created outside them, in messages, attachments and short decisions between people.
Pelenoir turns those traces into operational project memory before the LLM enters the scene.
Who waited. Who promised. Where approval is stuck. Which topic keeps returning. Where normal coordination becomes risk.
The manager does not read hundreds of messages. They see links, obligations and exceptions that can be filtered, verified and used.
Company memory
When a person leaves, the project must not lose context.
When a project manager leaves, the organisation often loses the map of shortcuts, conflicts, promises, exceptions and unwritten rules.
Pelenoir builds that map continuously from threads, attachments, contacts, roles, deadlines, projects and meaningful sentences.
Memory is not an archive. It is a structure you can filter, calculate, compare and audit.
That is why the foundation is a local database, graphs, FAISS indices, ontology, metrics and a controlled language model.
Know-how has value only when it survives a team change.
Layers
First data. Then indices and computation. Only then interpretation.
Enrichers and graphs
- Body segmentation
- Signatures and spam
- People timeline
- Social graph
- Project graph audit
Communication becomes relationships, roles, flow direction, projects and auditable structure.
Indices and labels
- FAISS
- Sentence store
- FIDIC labels
- RFI
- Approvals and escalations
The system can search both exactly and semantically. Every output can point back to its source.
Metrics and outputs
- Project Stress Index
- Weekly reports
- Benchmarks
- Audit log
- Controlled LLM
The model interprets only over prepared context. Truth remains in the data.
What Pelenoir does
Company communication becomes a management product.
Weekly project reports
Project Stress Index
People and flow metrics
Relationship graphs
Searchable thread trail
FIDIC sentence labels
Chat over controlled context
Deploy locally. Connect accounts and archives. Let the pipeline run ingest, enrichers, analytics, vectors, labels and metrics. The result is a view with provenance, filters and references.
Why AEC
Construction is the perfect test because silence is expensive here.
In construction, a problem rarely begins as a problem.
First an unanswered question hangs in the air. Then someone waits for a drawing. Then a claim arrives with a lawyer attached.
AEC is the right starting point: many parties, many obligations, many documents and a lot of expensive silence.
IFC, FIDIC and company ontology give the system vocabulary. Communication gives it behaviour.
When it works in AEC, the same principle extends into regulated enterprise, infrastructure, public administration and law.
A company that remembers does not need to investigate its own past.
Pelenoir turns project communication into local, auditable memory. Fewer impressions and less AI theatre. More questions that can be answered from data.