Develop FiftyOne Plugins

TL;DR — Read This First

Step 1: Decide what you're building.

Goal	Type	Read next
Run a computation, process samples, trigger actions	Operator	PYTHON-OPERATOR.md
Persistent panel in the grid, simple/form UI	Python Panel	PYTHON-PANEL.md
Persistent panel in the grid, rich/interactive UI	Hybrid Panel	HYBRID-PLUGINS.md + JAVASCRIPT-PANEL.md
Panel that opens inside the sample modal	JS Modal Panel	JAVASCRIPT-PANEL.md

Step 2: Read that file NOW — before generating any code. Each file contains patterns that are not in SKILL.md. Missing them causes the most common failures (wrong default= usage, handler binding errors, panel not appearing in App).

If something isn't rendering or you hit an error: Read TROUBLESHOOTING.md for the one-shot diagnostic command, reload vs restart table, and the top 5 error patterns with exact fixes.

For input types, Panel UI components reference, and a minimal working example: Read QUICK-REFERENCE.md.

Step 3: Clone the official plugins repo for reference patterns:

git clone https://github.com/voxel51/fiftyone-plugins.git /tmp/fiftyone-plugins 2>/dev/null || true

Step 4: Install your plugin by placing it (or symlinking it) in the plugins directory:

PLUGINS_DIR=$(python -c "import fiftyone as fo; print(fo.config.plugins_dir)")
# Convention: use underscores in the directory name (e.g. myorg_my_plugin) for shell compatibility
# The @org/name URI lives only inside fiftyone.yml — the directory name doesn't need to match
ln -s /path/to/my_plugin "$PLUGINS_DIR/my_plugin"

Step 5: Launch in debug mode so you see all server logs:

fiftyone app debug <dataset-name>

---

Key Directives

ALWAYS follow these rules:

1. Understand before coding

Ask clarifying questions. Never assume what the plugin should do.

2. Plan before implementing

Present file structure and design. Get user approval before generating code.

3. Search existing plugins for patterns

# Clone official plugins for reference
git clone https://github.com/voxel51/fiftyone-plugins.git /tmp/fiftyone-plugins 2>/dev/null || true

# Search for similar patterns
grep -r "keyword" /tmp/fiftyone-plugins/plugins/ --include="*.py" -l

list_plugins(enabled=True)
list_operators(builtin_only=False)
get_operator_schema(operator_uri="@voxel51/brain/compute_similarity")

4. Test locally before done

# Get plugins directory
PLUGINS_DIR=$(python -c "import fiftyone as fo; print(fo.config.plugins_dir)")

# Develop plugin in plugins directory
cd $PLUGINS_DIR/my-plugin

Write tests:

• Python: pytest for operators/panels
• JavaScript: vitest for React components

Verify in FiftyOne App before done.

5. Test complex logic outside the plugin first

If the plugin wraps non-trivial processing (model inference, image transforms, external APIs), verify that logic works as standalone Python before integrating it into the plugin:

# Test independently first
result = my_processing_fn(sample_path)
print(result)  # confirm correctness

# Only then wrap in execute()
def execute(self, ctx):
    result = my_processing_fn(ctx.dataset.first().filepath)
    ...

This prevents debugging two systems (plugin framework + custom logic) at once.

6. Iterate on feedback

Run server separately to see logs:

# Terminal 1: Python logs
python -m fiftyone.server.main

# Terminal 2: Browser at localhost:5151 (JS logs in DevTools console)

fiftyone app debug          # logs printed to shell; use with a dataset:
fiftyone app debug <dataset-name>

For automated iteration, use Playwright e2e tests:

npx playwright test

Refine until the plugin works as expected.

Critical Patterns

Gotcha 1 — default= must never reference state (causes infinite render loops)

# WRONG — triggers a render loop: schema changes → re-render → schema changes → ...
def render(self, ctx):
    panel = types.Object()
    panel.str("threshold", default=ctx.panel.get_state("threshold"))  # ← NEVER do this

# CORRECT — pass a literal; read state separately for display logic
def render(self, ctx):
    panel = types.Object()
    panel.str("threshold", default="0.5")  # literal only
    current = ctx.panel.get_state("threshold", "0.5")  # read separately if needed

The default= value is part of the schema the frontend uses to decide whether to re-render. If it changes between renders (because state changed), the frontend re-dispatches, which re-renders, infinitely.

Gotcha 2 — Event handlers must be bound instance methods, not static methods

# WRONG — @staticmethod strips __self__, breaking the handler URI silently
@staticmethod
def on_click(ctx):
    ...

panel.btn("btn", label="Go", on_click=MyPanel.on_click)  # ← broken

# CORRECT — regular instance method
def on_click(self, ctx):
    ...

panel.btn("btn", label="Go", on_click=self.on_click)  # ← works

The framework reads handler.__self__.uri to build the operator-trigger URI. A @staticmethod has no __self__, so the wiring silently fails and button clicks do nothing.

Gotcha 3 — Missing register(p) is a silent failure

# WRONG — plugin loads, no error, but ZERO operators appear in the App
import fiftyone.operators as foo

class MyPanel(foo.Panel):
    ...

# forgot register() entirely — or it raises an exception silently

# CORRECT — must explicitly register every operator and panel
def register(p):
    p.register(MyPanel)
    p.register(MyOperator)  # each class must be listed

The plugin loader calls module.register(self) after importing __init__.py. If the function is missing, raises an exception, or simply doesn't call p.register(YourClass) for a class, that class is never inserted into the operator registry. The App shows the plugin as enabled in list_plugins() — no error, no warning — but the operator or panel is completely invisible. This is the most common cause of "my panel doesn't appear in the + New panel menu."

Convention on directory naming: Use myorg_my_plugin (underscores, no @ or -) for clarity and shell compatibility. This is convention, not a hard requirement — the loader derives the module name from fiftyone.yml's name: field and sanitizes it automatically. The @org/name URI belongs in fiftyone.yml; the directory just needs to contain that file.

Operator Execution

# Chain operators (non-delegated operators only, in execute() only, fire-and-forget)
ctx.trigger("@plugin/other_operator", params={...})

# UI operations
ctx.ops.notify("Done!")
ctx.ops.set_progress(progress=0.5)  # keyword required — first positional arg is label, not progress

View Selection

# Use ctx.target_view() to respect user's current selection and filters
view = ctx.target_view()

# ctx.dataset - Full dataset (use when explicitly exporting all)
# ctx.view - Filtered view (use for read-only operations)
# ctx.target_view() - Filtered + selected samples (use for exports/processing)

Store Keys (Avoid Collisions)

# ctx.store() already scopes by dataset_id internally — no need to embed dataset name.
# Use a plugin-unique name to avoid collisions with other plugins on the same dataset.
def _get_store_key(self):
    return self.config.name.split("/")[-1]  # e.g. "my_panel"

store = ctx.store(self._get_store_key())

Panel State vs Execution Store

# ctx.panel.state - Transient (resets when panel reloads)
# ctx.store() - Persistent (survives across sessions)

def on_load(self, ctx):
    ctx.panel.state.selected_tab = "overview"  # Transient
    store = ctx.store(self._get_store_key())
    ctx.panel.state.config = store.get("user_config") or {}  # Persistent

Delegated Execution

Use for operations that: process >100 samples or take >1 second.

@property
def config(self):
    return foo.OperatorConfig(
        name="heavy_operator",
        allow_delegated_execution=True,
        default_choice_to_delegated=True,
    )

Progress Reporting

@property
def config(self):
    return foo.OperatorConfig(
        name="progress_operator",
        execute_as_generator=True,
    )

def execute(self, ctx):
    total = len(ctx.target_view())
    for i, sample in enumerate(ctx.target_view()):
        # Process sample...
        yield ctx.trigger("set_progress", {"progress": (i+1)/total})
    yield {"status": "complete"}

Custom Runs (Auditability)

Use Custom Runs for operations needing reproducibility and history tracking:

# Create run key (must be valid Python identifier - use underscores, not slashes)
run_key = f"my_plugin_{self.config.name}_v1_{timestamp}"

# Initialize and register
run_config = ctx.dataset.init_run(operator=self.config.name, params=ctx.params)
ctx.dataset.register_run(run_key, run_config)

See PYTHON-OPERATOR.md for full Custom Runs pattern. See EXECUTION-STORE.md for advanced caching patterns. See HYBRID-PLUGINS.md for Python + JavaScript communication.

Workflow

Phase 1: Requirements

Understand what the user needs to accomplish:

1. "What problem are you trying to solve?"
2. "What should the user be able to do?" (user's perspective)
3. "What information does the user provide?"
4. "What result does the user expect to see?"
5. "Any external data sources or services involved?"
6. "How will this fit into the user's workflow?"

Phase 2: Design

1. Search existing plugins for similar patterns
2. Choose the right panel architecture:
   • Modal panel (appears in sample modal) → Use JS Panel + Python Operators. See JAVASCRIPT-PANEL.md — Modal Panels. Do NOT use composite_view=True — it produces "Unsupported View" errors for modal panels.
   • Grid panel with rich UI → Use hybrid (Python + JavaScript). See HYBRID-PLUGINS.md.
   • Grid panel with simple UI → Use Python-only. See PYTHON-PANEL.md.
3. Create plan with:
   • Plugin name (@org/plugin-name)
   • File structure
   • Operator/panel specs
   • Input/output definitions
4. Get user approval before coding

See PLUGIN-STRUCTURE.md for file formats.

Phase 3: Generate Code

Before writing any code, read the reference file for your plugin type:

Building…	Read this file first
Operator	PYTHON-OPERATOR.md — input types, execution patterns, placement
Python panel	PYTHON-PANEL.md — layout containers, state/data/store, event lifecycle
JS/hybrid panel	JAVASCRIPT-PANEL.md + HYBRID-PLUGINS.md — React components, Python↔JS bridge
Persistent storage	EXECUTION-STORE.md — TTL caching, cross-session state
Something not rendering / errors	TROUBLESHOOTING.md — one-shot diagnostic, top 5 errors, reload vs restart
Input types, Panel UI components, minimal example	QUICK-REFERENCE.md — lookup tables and copy-paste starting point

Create these files:

File	Required	Purpose
fiftyone.yml	Yes	Plugin manifest
__init__.py	Yes	Python operators/panels
requirements.txt	If deps	Python dependencies
package.json	JS only	Node.js metadata
src/index.tsx	JS only	React components

For JavaScript panels with rich UI: Invoke the fiftyone-voodo-design skill for VOODO components (buttons, inputs, toasts, design tokens). VOODO is FiftyOne's official React component library.

Phase 4: Validate & Test

4.1 Validate Detection

list_plugins(enabled=True)  # Should show your plugin
list_operators()  # Should show your operators

If not found: Check fiftyone.yml syntax, Python syntax errors, restart App.

4.2 Validate Schema

get_operator_schema(operator_uri="@myorg/my-operator")

Verify inputs/outputs match your expectations.

4.3 Test Execution

set_context(dataset_name="test-dataset")
launch_app()
execute_operator(operator_uri="@myorg/my-operator", params={...})

Common failures:

• "Operator not found" → Check fiftyone.yml operators list
• "Missing parameter" → Check resolve_input() required fields
• "Execution error" → Check execute() implementation

Phase 5: Iterate

1. Get user feedback
2. Fix issues (sync source and plugins directory if separate)
3. Restart App if needed
4. Repeat until working

Quick Reference

For the full input types table, Panel UI components, config option tables, and a minimal working example:

→ Read QUICK-REFERENCE.md now.

Troubleshooting

One-shot plugin load diagnostic

Run this first whenever a plugin or operator is missing from the App:

curl -s -X POST http://localhost:5151/operators \
  -H 'Content-Type: application/json' -d '{}' \
  | python -c "
import json, sys
d = json.load(sys.stdin)
errors = d.get('errors', [])
if errors:
    print('PLUGIN LOAD ERRORS:')
    for e in errors: print(' -', e)
else:
    print('No load errors. Operators found:', len(d.get('operators', [])))
"

This returns every error the plugin loader captured — Python syntax errors, import failures, missing register() calls — in one shot. Check this before debugging anything else.

Plugin not appearing:

• Check fiftyone.yml exists in plugin root
• Verify location: ~/.fiftyone/plugins/
• Check register(p) is present and calls p.register() for each class — see Gotcha 3 above
• Check for Python syntax errors (run the diagnostic above)
• Restart FiftyOne App

Panel not appearing in the "+ New panel" menu:

• Plugin must load cleanly (check diagnostic above)
• PanelConfig.name must match the entry under panels: in fiftyone.yml
• Restart App after adding a new panel — the panel registry is built at startup

Operator not found:

• Verify operator listed in fiftyone.yml
• Check register() function
• Run list_operators() to debug

Secrets not available:

• Add to fiftyone.yml under secrets:
• Set environment variables before starting FiftyOne

Render loop / thousands of operator calls per minute:

• Almost always caused by default=ctx.panel.get_state(...) — see Gotcha 1 above
• Fix: replace with a string/number literal in default=

Advanced

Programmatic Operator Execution

# For executing operators outside of FiftyOne App context
import fiftyone.operators as foo
result = foo.execute_operator(operator_uri, ctx, **params)

Resources

• Plugin Development Guide
• Developing Panels
• Developing JS Plugins
• Panel Examples (reference implementations)
• FiftyOne Plugins Repo
• Operator Types API