Twelve Projects

An analytics workspace that investigates a question, runs the analysis, and hands back the work.

• Works across sources — Spreadsheets, shared files, SQL databases, market data, and Eidonix Lens knowledge catalogs.
• Runs the analysis — SQL, DuckDB, Python, R, and charting inside a controlled sandbox.
• Produces real artifacts — Charts, native Excel workbooks, and reports you can hand to someone.
• Carries knowledge forward — Skills leave with the retiring workforce. Cletus helps you capture and transfer that knowledge to the new generation.

Turns private documents and selected public-web material into a knowledge base that gets better as people use it.

• Catalogs — Manuals, filings, PDFs, images, and folders. Extraction and OCR run in the background.
• Web probes — Finds candidate sites, grades them, and crawls the ones worth keeping.
• Search — Hybrid retrieval with reranking scores, source links, and streamed summaries.
• API access — Scope application keys to specific catalogs or workspaces.

One endpoint for every model your tools need.

• Protocol translation — Anthropic Messages, OpenAI Responses, Chat Completions, and Bedrock Converse, translated automatically so any client reaches any backend.
• Model multiplexing — Local GPU servers, cloud APIs, and third-party providers behind a single model list.
• Documents and images — Text extraction from PDFs with an OCR fallback for scans. Images described for models that cannot see.
• Search and MCP — Web searches run server-side and the results are injected, with an MCP endpoint for agent clients.
• Key management — Proxy keys carry per-key model restrictions. Backend credentials never leave the server.

Analytics on top of eMaint X4 maintenance data.

• Syncs the source — Work orders, charges, PM schedules, and equipment records pulled from the eMaint API.
• Scores the fleet — Health scoring and KPI trending across sites and over time.
• Captures what was already solved — Every historic work order and the fix that closed it stays searchable in plain language. A technician describes the symptom and finds the last three times somebody dealt with it, and what worked.
• Reviews PMs — AI-assisted review of preventive maintenance schedules.
• Handles multiple sites — Per-user site access and a managed sync queue.

Live line status on wall-mounted displays.

• Screen designer — Cells, presses, buffers, arrows, and headers laid out on a full HD canvas.
• Live machine state — Decoded from PLC bitfields and streamed from WarLink.
• Shift aware — Planned minutes per hour, so breaks and planned downtime do not read as faults.
• Runs standalone — Each station runs on its own, with backup and restore from S3 or a central server.

Automated material transport between storage and the line.

• Request by type — Operators ask for what they need. The system picks the bin, routes a robot, and tracks delivery.
• FIFO by default — The oldest eligible material always goes first, enforced by the system.
• Chain of custody — Every bin carries a QR code scanned at pickup to confirm identity.
• Survives the network — Edge stations keep tracking local state and queue messages when the link to core drops.
• Vendor independent — The robot fleet sits behind an interface, so changing vendor does not touch dispatch.

Vibration and condition monitoring for rotating equipment.

• Collects continuously — Sensor data from plant equipment into object storage.
• Keeps trends readable — Long time series reduced and aggregated so months of data stay legible.
• Warns early — Developing faults surfaced while the repair can still be scheduled.

Prints markdown straight to thermal receipt printers.

• Real formatting — Headings, bold, underline, tables, images, QR codes, paper cuts, and the buzzer, from a plain .md file.
• Any connection — TCP, USB, or serial, with any ESC/POS printer.
• Tunable output — Paper width and line density configurable per job.
• Scriptable — Reads a file or stdin, so it drops into an existing work order flow without glue code.

Structured reporting that rolls up without losing the original.

• Submit, do not email — People submit to a manager they select. Each level consolidates what is beneath it.
• No telephone effect — Any summary line drills back to the submission it came from.
• Themes across the org — Recurring topics extracted across submissions and attributed to author, unit, and period.
• Scope enforced — Visibility is checked on every read, not just on lists.

Data Matrix code generation for part marking.

• Standard sizes — ASCII payloads encoded across the standard Data Matrix size range.
• Correct by construction — Code word capacity and error correction layout handled per size.
• Small — Dependency-light, so it embeds in a labelling or marking step.