M5 — @thinkfleet/piece-agentmark Activepieces piece

pieces/agentmark/README.md

@@ -0,0 +1,46 @@
+# @thinkfleet/piece-agentmark
+
+[Activepieces](https://www.activepieces.com/) piece for [AgentMark](https://github.com/ThinkfleetAI/agentmark) — convert any web page or PDF into a compact AgentMark snapshot from inside a flow, and fill PDF forms with data from previous flow steps.
+
+## What this piece adds
+
+| Action | What it does | Inputs | Outputs |
+|---|---|---|---|
+| **Capture Web Page** | Launch Chromium, snapshot a URL, return AgentMark | `url`, `wait_until`, `timeout_ms`, `headless` | `agentmark`, `url`, `title`, `kind`, `action_count`, `bytes` |
+| **Capture PDF** | Convert a PDF (URL/file/base64/data URI) to AgentMark; optional OCR | `source`, `source_url`, `title`, `password`, `enable_ocr`, `ocr_language` | `agentmark`, `source_url`, `bytes`, `ocr_used` |
+| **Fill PDF Form** | Fill an AcroForm PDF and return the filled bytes | `source`, `values`, `flatten`, `return_format`, `password` | `filled_pdf` (data URI or raw base64), `bytes`, `fields_applied`, `fields_skipped`, `flattened` |
+
+All actions run on the Activepieces worker — no external service required. Browser-based snapshots use Chromium via Playwright; PDF support uses pdfjs-dist + pdf-lib; optional OCR uses Tesseract.js with Poppler (`pdftoppm`).
+
+## Why use this in a flow
+
+- **Drop AgentMark into any agent flow** without writing code. The agent loop lives in your flow — call `Capture Page`, pass the snapshot to your AI step, then `Page Execute` (coming soon) or chain another snapshot.
+- **Fill insurance/government/vendor PDFs** from CRM data. Map field action IDs from a previous step to records pulled from your data store, run `Fill PDF Form`, attach the result to an email.
+- **Extract structured data from scanned docs** by enabling OCR on `Capture PDF`. Works on "Microsoft Print To PDF" output and scanner outputs.
+
+## Activepieces setup
+
+This piece depends on:
+
+- `@thinkfleet/agentmark` (the core library)
+- `playwright-core` for browser-based actions
+- `pdfjs-dist` (optional, required for any PDF action)
+- `pdf-lib` (optional, required for `Fill PDF Form`)
+- `tesseract.js` + Poppler installed on the worker (optional, required for OCR)
+
+Install Chromium binaries on the worker once:
+
+```bash
+npx playwright-core install chromium
+```
+
+Install Poppler on the worker (only if using OCR):
+
+```bash
+brew install poppler           # macOS
+apt-get install poppler-utils  # Ubuntu/Debian
+```
+
+## License
+
+MIT.

pieces/agentmark/package.json

@@ -0,0 +1,61 @@
+{
+    "name": "@thinkfleet/piece-agentmark",
+    "version": "0.1.0",
+    "description": "Activepieces piece — convert any web page or PDF into an AgentMark snapshot, then drive it from any flow.",
+    "type": "commonjs",
+    "main": "./dist/src/index.js",
+    "types": "./dist/src/index.d.ts",
+    "license": "MIT",
+    "repository": {
+        "type": "git",
+        "url": "https://github.com/ThinkfleetAI/agentmark.git",
+        "directory": "pieces/agentmark"
+    },
+    "homepage": "https://agentmark.dev",
+    "keywords": [
+        "activepieces",
+        "piece",
+        "agentmark",
+        "ai",
+        "browser",
+        "pdf",
+        "ocr",
+        "form-fill"
+    ],
+    "scripts": {
+        "build": "tsc -p tsconfig.lib.json && cp package.json dist/",
+        "test": "vitest run"
+    },
+    "files": [
+        "dist",
+        "README.md"
+    ],
+    "dependencies": {
+        "@thinkfleet/agentmark": "file:../..",
+        "tslib": "^2.3.0",
+        "undici": "^7.0.0"
+    },
+    "_publishing_note": "@thinkfleet/agentmark is `file:../..` for in-monorepo development; bump to ^0.7.0 (or whatever version is on npm) before publishing this piece.",
+    "peerDependencies": {
+        "@activepieces/pieces-framework": ">=0.7.0",
+        "playwright-core": ">=1.40.0"
+    },
+    "peerDependenciesMeta": {
+        "@activepieces/pieces-framework": {
+            "optional": false
+        },
+        "playwright-core": {
+            "optional": false
+        }
+    },
+    "devDependencies": {
+        "@activepieces/pieces-framework": "*",
+        "@types/node": "20.19.9",
+        "pdf-lib": "^1.17.1",
+        "pdfjs-dist": "^4.10.38",
+        "playwright-core": "^1.40.0",
+        "tesseract.js": "^5.1.1",
+        "typescript": "^5.4.0",
+        "vitest": "3.0.8"
+    }
+}

pieces/agentmark/src/index.ts

@@ -0,0 +1,27 @@
+/**
+ * @thinkfleet/piece-agentmark — Activepieces piece for AgentMark.
+ *
+ * Drop into any Activepieces flow to convert web pages or PDFs into
+ * AgentMark snapshots and fill PDF forms — all without leaving the flow
+ * builder. Wraps the core @thinkfleet/agentmark library.
+ */
+
+import { createPiece, PieceAuth } from '@activepieces/pieces-framework'
+import { snapshotWebPage } from './lib/actions/snapshot-web-page'
+import { snapshotPdf } from './lib/actions/snapshot-pdf'
+import { fillPdfForm } from './lib/actions/fill-pdf-form'
+
+export const agentmark = createPiece({
+    displayName: 'AgentMark',
+    description:
+        'Convert web pages and PDFs into compact AgentMark snapshots; fill '
+        + 'AcroForm PDFs from flow data. Powered by @thinkfleet/agentmark.',
+    auth: PieceAuth.None(),
+    minimumSupportedRelease: '0.78.0',
+    logoUrl: 'https://agentmark.dev/logo.svg',
+    authors: ['thinkfleet'],
+    actions: [snapshotWebPage, snapshotPdf, fillPdfForm],
+    triggers: [],
+})
+
+export { snapshotWebPage, snapshotPdf, fillPdfForm }

pieces/agentmark/src/lib/actions/fill-pdf-form.ts

@@ -0,0 +1,105 @@
+import { createAction, Property } from '@activepieces/pieces-framework'
+import { openPdfDocument } from '@thinkfleet/agentmark'
+import { resolveBytes, bytesToBase64DataUri } from '../common'
+
+export const fillPdfForm = createAction({
+    name: 'fill_pdf_form',
+    displayName: 'Fill PDF Form',
+    description:
+        'Fill an AcroForm PDF in one atomic step. Pass a values object keyed '
+        + 'by AgentMark action ID OR by original field name; the action '
+        + 'matches either. Returns the filled PDF as a base64 data URI.',
+    props: {
+        source: Property.LongText({
+            displayName: 'PDF Source',
+            description:
+                'HTTP(S) URL, file path, file:// URI, data: URI, or base64 string.',
+            required: true,
+        }),
+        values: Property.Json({
+            displayName: 'Field Values',
+            description:
+                'Object mapping field IDs (action IDs like `act_field_1`) or '
+                + 'field names (e.g. `applicant.first_name`) to values. '
+                + 'Strings for text/select/radio, booleans for checkboxes, '
+                + 'arrays for multi-select.',
+            required: true,
+            defaultValue: {},
+        }),
+        flatten: Property.Checkbox({
+            displayName: 'Flatten',
+            description:
+                'Bake values into page content. Resulting PDF is no longer fillable.',
+            required: false,
+            defaultValue: false,
+        }),
+        return_format: Property.StaticDropdown({
+            displayName: 'Return Format',
+            description: 'How the filled PDF is returned in the action output.',
+            required: false,
+            defaultValue: 'data_uri',
+            options: {
+                disabled: false,
+                options: [
+                    { label: 'Base64 data URI', value: 'data_uri' },
+                    { label: 'Raw base64 (no scheme)', value: 'base64' },
+                ],
+            },
+        }),
+        password: Property.ShortText({
+            displayName: 'Password',
+            description: 'Password for encrypted PDFs.',
+            required: false,
+        }),
+    },
+    async run(context) {
+        const { source, values, flatten, return_format, password } = context.propsValue
+        const data = await resolveBytes(source)
+        const doc = await openPdfDocument({
+            data,
+            sourceUrl: source.startsWith('http') ? source : 'inline:pdf',
+            password,
+        })
+
+        try {
+            const valuesMap = (values ?? {}) as Record<string, unknown>
+
+            // Build action-id-keyed dispatch map from BOTH action IDs and
+            // original field names. Caller can use whichever is convenient.
+            const actionIdByName = new Map<string, string>()
+            for (const [actionId, field] of doc.fields) {
+                actionIdByName.set(field.fieldName, actionId)
+            }
+
+            const summary: Array<{ key: string; resolved_action_id: string }> = []
+            const skipped: string[] = []
+
+            for (const [key, value] of Object.entries(valuesMap)) {
+                const resolved = doc.fields.has(key)
+                    ? key
+                    : actionIdByName.get(key)
+                if (!resolved) {
+                    skipped.push(key)
+                    continue
+                }
+                await doc.execute(resolved, value)
+                summary.push({ key, resolved_action_id: resolved })
+            }
+
+            const filled = await doc.save({ flatten: flatten === true })
+            const out = return_format === 'base64'
+                ? Buffer.from(filled).toString('base64')
+                : bytesToBase64DataUri(filled)
+
+            return {
+                filled_pdf: out,
+                bytes: filled.length,
+                fields_applied: summary,
+                fields_skipped: skipped,
+                flattened: flatten === true,
+            }
+        } finally {
+            await doc.close()
+        }
+    },
+})

pieces/agentmark/src/lib/actions/snapshot-pdf.ts

@@ -0,0 +1,101 @@
+import { createAction, Property } from '@activepieces/pieces-framework'
+import {
+    convertPdf,
+    PopplerRenderBackend,
+    TesseractOcrBackend,
+} from '@thinkfleet/agentmark'
+import { resolveBytes } from '../common'
+
+export const snapshotPdf = createAction({
+    name: 'snapshot_pdf',
+    displayName: 'Capture PDF',
+    description:
+        'Convert a PDF (URL, file path, base64, or data URI) into a compact '
+        + 'AgentMark snapshot. PDFs with form fields produce kind: \'form\'; '
+        + 'plain documents produce kind: \'document\'. Optionally OCR pages '
+        + 'with no extractable text using Tesseract + Poppler.',
+    props: {
+        source: Property.LongText({
+            displayName: 'Source',
+            description:
+                'HTTP(S) URL, file path, file:// URI, data:application/pdf;base64,... '
+                + 'URI, or a bare base64 string.',
+            required: true,
+        }),
+        source_url: Property.ShortText({
+            displayName: 'Source URL (override)',
+            description:
+                'Optional URI to record as the snapshot\'s `url` field. Useful '
+                + 'when the input is a data URI or in-memory base64 and you '
+                + 'want a stable identifier for downstream steps.',
+            required: false,
+        }),
+        title: Property.ShortText({
+            displayName: 'Title (override)',
+            description: 'Override the document title. Leave blank to use the PDF metadata title.',
+            required: false,
+        }),
+        password: Property.ShortText({
+            displayName: 'Password',
+            description: 'Password for encrypted PDFs.',
+            required: false,
+        }),
+        enable_ocr: Property.Checkbox({
+            displayName: 'Enable OCR',
+            description:
+                'Run Tesseract OCR on pages with no extractable text. Required '
+                + 'for scanned PDFs and "Microsoft Print To PDF" output. Slower; '
+                + 'requires Poppler installed on the worker host (pdftoppm).',
+            required: false,
+            defaultValue: false,
+        }),
+        ocr_language: Property.ShortText({
+            displayName: 'OCR Language',
+            description: 'BCP-47 language hint. Default: eng.',
+            required: false,
+            defaultValue: 'eng',
+        }),
+    },
+    async run(context) {
+        const {
+            source,
+            source_url,
+            title,
+            password,
+            enable_ocr,
+            ocr_language,
+        } = context.propsValue
+
+        const data = await resolveBytes(source)
+        const sourceUrl = source_url
+            ?? (source.startsWith('http') ? source : 'inline:pdf')
+
+        const ocrBackend = enable_ocr ? new TesseractOcrBackend({ language: ocr_language ?? 'eng' }) : undefined
+        try {
+            const { agentmark } = await convertPdf({
+                data,
+                sourceUrl,
+                title,
+                password,
+                ocr: enable_ocr
+                    ? {
+                        render: new PopplerRenderBackend(),
+                        ocr: ocrBackend!,
+                        mode: 'auto',
+                        dpi: 200,
+                        language: ocr_language ?? 'eng',
+                    }
+                    : undefined,
+            })
+
+            return {
+                agentmark,
+                source_url: sourceUrl,
+                bytes: agentmark.length,
+                ocr_used: enable_ocr === true,
+            }
+        } finally {
+            await ocrBackend?.close().catch(() => {})
+        }
+    },
+})