Available today, in development, ahead.
A roadmap is only useful if it separates work you can use from work we intend to do. This page shows the platform expanding outward from one data layer, and publishes no dates.
Four moves, in order.
Each move only makes sense standing on the one before it, and each one says plainly where it stands. Everything here is on the map below, sorted by when it arrives instead of by what it is.
-
Start from what already runs.
SQL, JSON and time series are in the platform today, along with the storage layout, transaction boundary and snapshot reads underneath them. The plan starts there rather than from a rewrite, every move after this one has to sit on the same engine, planner and storage contract.
- Relational / SQL
- JSON / Document
- Time-series
- Row storage
- Columnar storage
- Transactions / ACID
- MVCC
In the platform today
-
Land the models already being built.
Vector, graph and blob data are being built now. Each one shares the parser, planner, execution and storage path that already exists, so it arrives as another way to read the same data rather than as another system to run beside it.
- Vector / embeddings
- Graph
- Blob / object data
In development · not available to use yet
-
Open the remaining paths to the same data.
Full-text, key-value, geospatial, event and ledger data are being explored as further access paths into data that already lives in the layer, not as separate stores to keep in step. Nothing in this move is scheduled.
- Full-text search
- Key-value
- Geospatial
- Event / stream data
- Ledger / immutable records
Being explored · not scheduled
-
Then take the layer outward.
Where the system runs, distribution, region boundaries, replication, more than one class of backing storage, and the enterprise systems a business already operates. Those systems stay where they are; the direction is to meet them, not to move them. No enterprise system is integrated today.
- Distribution
- Multi-region
- Replication
- Multi-storage
- Multimodal data
- AI memory
Design open · not scheduled
Same layer · same storage contract · more ways to reach the data
Rule If a capability is not on the map below, it is not available to use, and nothing on this site will describe it as though it were. A move is a direction the layer is being taken in, not a date, and not a delivery promise.
Three words, used the same way everywhere.
Available
Shipped and in the product.
In development
Being built now, and not available to use yet.
Roadmap
Longer-term roadmap. Not scheduled, and not available to use.
The whole map, direction by direction.
Five tiers, one legend. Read the frame, not a repeated badge: solid is usable today, dashed is being built, dotted is potential, and the faintest frame is long-term.
- Current Usable today.
- Current foundation Usable today, and the groundwork later capabilities build on.
- Roadmap · in development Being built now. Not available to use yet.
- Potential Being explored. Not scheduled, and nothing to build on yet.
- Long-term A direction, further out. Not scheduled.
| Data type / model | PLOMID direction | Example |
|---|---|---|
| Relational / SQL | Tables, keys, joins and transactions | users, orders, payments |
| JSON / Document | Nested objects, queried alongside rows | profiles, configurations, API data |
| Time-series | Events and measures with the data they describe | telemetry, metrics, events |
| Row storage | Page-oriented layout for mixed shapes | transactional workloads |
| Columnar storage | Ordered reads for aggregate work | analytics |
| Transactions / ACID | One boundary across every data model | financial / operational transactions |
| MVCC | Readers and writers that do not block each other | concurrent transactions |
| Vector / embeddings | Similarity as another access path | semantic search, RAG |
| Graph | Relationships as first-class structures | relationships, knowledge graphs |
| Blob / object data | Large objects with queryable metadata | images, video, PDFs, datasets |
| Full-text search | Relevance over text already in the layer | documents, logs, knowledge bases |
| Key-value | Fast lookups for short-lived state | sessions, state, caching |
| Geospatial | Location-aware predicates and ranges | maps, vehicles, logistics |
| Event / stream data | Continuous input read as stored data | IoT, telemetry, event systems |
| Ledger / immutable records | Append-only history that can be audited | audit trails, financial records |
| Multimodal data | One layer across text, image, audio and vectors | text + image + audio + video + vectors |
| AI memory | Persistent, queryable memory for agents | persistent agent / user memory |
Reading it Current and current foundation are usable today. Everything below them is direction: in development means being built right now, potential means being explored, and long-term means further out. None of the three is scheduled, and nothing on this page will present them as shipped.
The stack, layer by layer.
A layer is marked roadmap when its design depends on decisions we have not finished making.
| Layer | Status | Responsibility |
|---|---|---|
| Interfaces | Available | The surfaces applications, AI clients and developers talk to. |
| Planner & coordination | Available | Turns a request into a plan over one data layer. |
| Execution | Available | Runs the plan close to where the data lives. |
| Storage | Available | Rows, documents and events persisted through one storage contract. |
| Deployment fabric | Roadmap | Where the system runs, and where data is allowed to be. |
On dates We do not publish delivery dates. A capability moves to available when it can be used and documented, and not before, so the label you see here is the current truth rather than a target.
How the layer may expand outward.
Data models decide what you can express. Infrastructure decides where the system runs and what a region means. This part of the design is open, and it is marked that way.
- Distribution
Running the same stack across more than one node, without changing the query surface.
Roadmap - Multi-region
Region boundaries that are declared policy rather than an accident of setup.
Roadmap - Replication
Where copies may exist, and what makes them consistent.
Roadmap - Multi-storage
One system spanning more than one class of backing storage.
Roadmap
One layer across the systems a business already runs.
Enterprises do not move their data into one place, and mostly should not. The direction is to meet those systems where they are: PLOMID as a common data layer over sources that stay where they are. Nothing below is integrated today.
5 source categories · 4 consuming surfaces · 0 integrated today
Sources stay where they are; the layer meets them. Select a system above to read what the direction means for it.
Read without moving
Query the system where it lives, through the layer. No copy, no pipeline to maintain.
DirectionWrite back where it belongs
Changes land in the system that owns them, under one transaction boundary.
DirectionOne answer, every surface
Applications, analytics and AI workloads read the same state, not three copies of it.
DirectionDirection Foreign tables, external data and enterprise connectors are the direction the layer is being taken in: PLOMID as a common data layer across the systems a business already runs. No enterprise system is integrated today, and no logo here implies otherwise — names are categories, drawn with the site's own marks.
The direction, stated as direction.
PLOMID already carries several data models in one layer. These are the directions that extend it further, more models, deeper integration, broader workload support, not a promise to become multi-model later. None is scheduled.
Extending the model layer
More shapes of data reachable through the same query surface, multimodal content across text, image, audio and video, treated as data rather than as files beside it.
Long-term · not scheduledAI memory
Persistent, queryable memory for agents, stored, versioned and read back like any other data, in the same layer as the records it refers to.
Long-term · not scheduledLedger records
Append-only history that can be audited without a separate system beside it.
Long-term · not scheduledWorkloads the layer is built to grow into.
A capability is only useful when it lands on a workload. These are the patterns the additions are aimed at — three that run today, and three that depend on work still being built.
Application data
Records that are structured, plus the fields that never quite fit a column.
- Rows for core records
- Documents for changing shape
- One statement across both
Real-time data
What is happening now, and what led here.
- Events and measurements
- Windowed reads
- No export between systems
Operational reporting
Answers asked of live data rather than a nightly copy.
- Aggregates in place
- Document fields filterable
- One copy of the truth
Search and retrieval
Find what is similar, not only what matches.
- Similarity search path
- Documents stay queryable
- One layer to keep in step
Connected data
What is reachable, and what depends on what.
- Nodes and edges as structures
- Traversal without rebuild
- Over data already in the layer
Large objects
Files, media and artifacts that need to be found by query.
- Big immutable objects
- Queryable metadata
- No separate catalogue
Follow the platform as it evolves.
A short note when something moves from direction into the platform. No dates promised, no newsletter filler, and nothing to unsubscribe from that you did not ask for.