DevAgent

AI-assisted industrial HMI engineering for Ignition Perspective

DevAgent turns customer plant information into reviewable 2D/3D HMI concepts, validated industrial engineering models, and reproducible native Ignition Perspective candidates.

Quick start · Architecture · Creator workflow · Ignition handoff · Verification

Current scope: engineering and isolated-lab candidates. Authenticated Gateway import, native Perspective runtime FAT, real OPC UA/PLC mapping validation, and production release have not been qualified. Production is HOLD.

Capabilities

Architecture

   CUSTOMER INPUTS
   Requirements | Equipment | Connections | PLC / tag exports
                           |
                           v
                  BYOM AI AGENT
                  Proposal + review
                           |
                           v
              ENGINEERING VALIDATION
         Equipment | Signals | Topology | Safety
                           |
            +--------------+---------------+
            |              |               |
            v              v               v
       2D HMI         3D HMI         Native Ignition
       preview        preview        project candidate
            |              |               |
            +--------------+---------------+
                           |
                           v
               ISOLATED LAB + OFFLINE FAT
                           |
                    Gateway import
                    and runtime tests
                    [NOT YET QUALIFIED]
                           |
                           v
              Real OPC UA source validation
              Independent FAT / site approval
                           |
                           v
                APPROVED PRODUCTION RELEASE
                         [HOLD]

Engineering rule: AI proposes; deterministic checks validate; simulation and a real test Gateway supply separate evidence; an authorized engineer approves. A synthetic PASS cannot establish physical PLC correctness.

Quick start

Install and run (Ubuntu / Linux)

Requires Python 3.13. Install the published package from PyPI, then launch the local Creator:

python3.13 -m pip install --no-cache-dir --index-url https://pypi.org/simple/ devagent-ignition==0.1.1
devagent start

Open http://127.0.0.1:8895/ in your browser. No GitHub clone is required. If port 8895 is in use, run devagent start --port 8962 and open http://127.0.0.1:8962/.

If devagent is not found, make sure the Python 3.13 environment you installed into is activated and its bin directory is on your PATH. Use a dedicated virtual environment if you also have devagent-ai installed, since both packages provide a devagent executable.

Develop or verify from source

Requirements: Linux/Ubuntu, Git, and Python 3.13. Optional browser and OPC UA tests have additional dependencies.

git clone https://github.com/tomha85/devagent-ignition.git
cd devagent-ignition

sha256sum -c SHA256SUMS.txt
python3 -m unittest discover -s tests -q
python3 devagent_hmi.py preflight

# Start the local HMI Creator on port 8895.
./start.sh

Open http://127.0.0.1:8895/ on the same computer.

Choose another local port if 8895 is occupied:

DEVAGENT_SPATIAL_PORT=8962 ./start.sh

The local Creator does not automatically connect to an Ignition Gateway or modify PLCs. Use python3 devagent_hmi.py --help to explore CLI commands.

Creator workflow

1. Connect AI (optional). Choose your own supported provider/model and test the connection. Review all generated drafts; AI responses are not verified engineering facts.
2. Describe the site. Enter plant requirements and provide available equipment, tag, and connection files. See sample customer inputs.
3. Choose a starting point. Select 2D Flow, 3D Flow, or New Project. The Original V8/R9 options are source-backed demos, not a representation of your plant.
4. Analyze and validate. Resolve contract errors and review any inferred or missing equipment and signal relationships.
5. Generate a preview. Navigate floors, select equipment, and inspect only available diagnostics.
6. Review evidence. Keep synthetic testing separate from real PLC data and qualify the application in an isolated Gateway before production consideration.

Validate a model without launching the UI:

python3 devagent_hmi.py validate --model examples/warehouse.json

Other included samples: single floor, warehouse, and water treatment.

BYOM and data handling

DevAgent supports multiple cloud and local provider adapters, including compatible Ollama/vLLM endpoints. A successful model connection test establishes connectivity, not engineering correctness. Follow your organization's rules before sending plant documents, tag exports, or operational data to an external provider.

See AI provider setup for connection details and security limitations.

Native Ignition handoff

The Native Ignition Perspective compiler generates an offline, read-only Perspective candidate. Not implemented: qualified one-click customer deployment and automatic creation of real OPC UA connections and Gateway tag providers. Gateway import and runtime behavior still require isolated qualification.

The engineering handoff is:

1. Obtain an authorized, genuine Ignition Perspective project export for the template envelope.
2. Validate the customer model and prepare an engineer-reviewed, lab-scoped binding manifest.
3. Compile and verify the native candidate against the reference ZIP and source hashes.
4. Import only into an approved, backed-up isolated test Gateway.
5. Verify actual Perspective rendering, floor/equipment interactions, tag quality, and OPC-to-HMI FAT.
6. Validate real source identity and site behavior independently before requesting production approval.

Create a pending binding manifest (tag paths are intentionally empty):

python3 devagent_hmi.py v07-binding-skeleton \
  --model examples/warehouse.json \
  --out /tmp/devagent-pending-lab-bindings.json

The manifest is not preapproved. An engineer must review its model identity, decisions, and any lab paths before compilation. Discover the exact compiler flags with:

python3 devagent_hmi.py v083-native-compile --help
python3 devagent_hmi.py v083-native-verify --help

Never infer OPC UA NodeIds, motor/VFD associations, PE IN/OUT semantics, speeds, fault states, or operating quality from a demo.

Read the native operator/deployment notes and native binding/FAT guide.

Verification

sha256sum -c SHA256SUMS.txt
python3 -m compileall -q devagent_spatial devagent_hmi.py tests
python3 -m unittest discover -s tests -q
python3 devagent_hmi.py preflight

GitHub Actions: .github/workflows/offline-ci.yml. Its CI jobs cover offline tests, static checks, a browser workflow, and isolated synthetic OPC UA tests. CI PASS is not a live Gateway FAT or site-acceptance result.

Current qualification boundaries

Unavailable values must show NOT MAPPED, UNKNOWN, or a clear BAD/STALE quality indication, not fabricated operating states. START/STOP/RESET actions remain disabled in the unqualified native operator candidate.

Repository guide

For engineering detail, see the qualification record, native import milestone, and production acceptance gates. Historical milestones remain in separate documentation; this README is the starting point.

DevAgent's purpose: faster, more repeatable industrial HMI engineering with explicit signal provenance, deterministic verification, and evidence-based commissioning.