Rossum Deploy

Minimalistic pull/diff/push deployment for Rossum configurations.

Note: This is a lightweight alternative to the deployment-manager (PRD2) tool. We aim to provide a simple, dependency-light solution for the rossum-agent that covers core deployment workflows.

Comparison with deployment-manager

Feature	rossum-deploy	deployment-manager (PRD2)
Object Types	9 core types	All 16 types ✅
Commands	pull, diff, push	pull, push, purge, deploy, init, hook (payload/test/sync), docommando, llm-chat
Multi-target Deploy	✅ (copy_org + deploy)	✅ (deploy to multiple orgs)
Attribute Override	❌	✅ (regex-based overrides)
3-way Merge	❌	✅ (conflict resolution)
GIT Integration	✅ (tracks local changes)	✅ (tracks staged/unstaged changes)
Subdirectories	❌	✅ (dev/uat/prod structure)
Secrets Management	❌	✅ (git-ignored secrets)
LLM Documentation	✅ (via rossum-agent)	✅ (docommando)
Dependencies	Minimal (pydantic, rossum-api)	Full (click, rich, questionary, etc.)
API	Python-first	CLI-first

Supported Object Types

rossum-deploy supports 9 core Rossum object types:

• Core: Workspace, Queue, Schema, Inbox
• Automation: Hook, Connector
• Engine & AI: Engine
• Templates: EmailTemplate
• Rules: Rule

Note: deployment-manager supports additional types (Organization, Annotation, Document, RuleTemplate, Workflow, WorkflowStep, Task) for production deployments.

When to use rossum-deploy:

• Simple pull/push workflows within an AI agent
• Minimal dependency footprint required
• Programmatic access from Python code

When to use deployment-manager:

• Production deployments across multiple organizations
• Complex attribute overrides and conflict resolution
• GIT-based workflow tracking
• CLI-driven operations

Installation

pip install -e rossum-deploy/

Usage

Python API

from rossum_deploy import Workspace

# Initialize workspace
ws = Workspace(
    "./my-project",
    api_base="https://api.elis.rossum.ai/v1",
    token="your-token"
)

# Pull all objects from an organization
result = ws.pull(org_id=123456)
print(result.summary())

# Show diff between local and remote
diff = ws.diff()
print(diff.summary())

# Push changes (dry run first)
result = ws.push(dry_run=True)
print(result.summary())

# Push for real
result = ws.push()
print(result.summary())

Full Example: Pull Org, Create Queue, and Add Hook

from rossum_deploy import Workspace
from rossum_api import SyncRossumAPIClient
from rossum_api.dtos import Token

# Initialize client and workspace
api_base = "https://api.elis.rossum.ai/v1"
token = "your-api-token"

client = SyncRossumAPIClient(api_base, Token(token))
ws = Workspace("./my-rossum-project", api_base=api_base, token=token)

# 1. Pull existing organization configuration
org_id = 123456
result = ws.pull(org_id=org_id)
print(result.summary())
# Output:
# # Pull Summary
# - Pulled: 15
# - Skipped: 0
# ## Pulled
# - workspace: Main Workspace (111111)
# - queue: Invoices (222222)
# ...

# 2. Create a new queue using the Rossum API client
workspace_url = f"{api_base}/workspaces/111111"
schema = client.create_schema(name="New Queue Schema", content=[
    {
        "category": "section",
        "id": "invoice_section",
        "label": "Invoice Details",
        "children": [
            {
                "category": "datapoint",
                "id": "invoice_number",
                "label": "Invoice Number",
                "type": "string",
                "rir_field_names": ["invoice_id"]
            },
            {
                "category": "datapoint",
                "id": "total_amount",
                "label": "Total Amount",
                "type": "number",
                "rir_field_names": ["amount_total"]
            }
        ]
    }
])

queue = client.create_queue(
    name="My New Queue",
    workspace=workspace_url,
    schema=f"{api_base}/schemas/{schema.id}"
)
print(f"Created queue: {queue.name} (ID: {queue.id})")

# 3. Create a hook and attach it to the queue
hook = client.create_hook(
    name="Invoice Validation Hook",
    type="function",
    queues=[f"{api_base}/queues/{queue.id}"],
    events=["annotation_content.initialize"],
    config={
        "code": '''
def rossum_hook_request_handler(payload):
    """Validate invoice data on initialization."""
    messages = []

    for field in payload.get("annotation", {}).get("content", []):
        if field.get("schema_id") == "invoice_number":
            if not field.get("value"):
                messages.append({
                    "id": field["id"],
                    "type": "warning",
                    "content": "Invoice number is required"
                })

    return {"messages": messages}
''',
        "runtime": "python3.12"
    },
    sideload=["annotation"]
)
print(f"Created hook: {hook.name} (ID: {hook.id})")

# 4. Pull again to sync local state with new objects
result = ws.pull(org_id=org_id)
print(result.summary())

# 5. Check diff to verify everything is in sync
diff = ws.diff()
print(diff.summary())
# Output:
# # Diff Summary
# - Unchanged: 17
# - Local modified: 0
# - Remote modified: 0
# - Conflicts: 0

CLI (Planned)

Note: CLI support is planned but not yet implemented. Use the Python API for now.

# Future CLI commands (not yet available):
# rossum-deploy pull 123456
# rossum-deploy diff
# rossum-deploy push --dry-run
# rossum-deploy push

Local File Structure

my-project/
├── workspaces/
│   └── {name}_{id}.json
├── queues/
│   └── {name}_{id}.json
├── schemas/
│   └── {name}_{id}.json
├── hooks/
│   └── {name}_{id}.json
├── inboxes/
│   └── {name}_{id}.json
└── .id_mapping_{source}_to_{target}.json  # Created by copy_org

Cross-Organization Deployment

rossum-deploy supports a safe workflow for deploying changes from a sandbox to production.

Step 1: Copy Production to Sandbox

from rossum_deploy import Workspace

ws = Workspace("./my-project", api_base="...", token="...")

# Copy production config to sandbox (one-time setup)
result = ws.copy_org(
    source_org_id=123456,  # Production
    target_org_id=789012,  # Sandbox
)
print(result.summary())

This creates all objects in sandbox and saves an ID mapping file.

Step 2: Agent Modifies Sandbox via API

The AI agent makes changes in sandbox using the Rossum API client.

Step 3: Pull Sandbox and Review

# Pull sandbox state
ws.pull(org_id=789012)

# Show diff
print(ws.diff().summary())

Step 4: Deploy to Production

# Dry run first
result = ws.deploy(target_org_id=123456, dry_run=True)
print(result.summary())

# Deploy for real after user approval
result = ws.deploy(target_org_id=123456)
print(result.summary())

Workspace Comparison (Agent Workflow)

This is the recommended workflow for AI agents to safely develop and test changes before deployment.

The Workflow

1. Copy a workspace to create an isolated sandbox
2. Modify the sandbox (agent makes changes via API)
3. Pull both original and modified workspaces
4. Compare to see exactly what changed

Step 1: Copy Workspace to Sandbox

from rossum_deploy import Workspace

ws = Workspace("./source-project", api_base="...", token="...")

# Copy a single workspace (and all its objects) to target org
result = ws.copy_workspace(
    source_workspace_id=111111,
    target_org_id=789012,  # Sandbox org
)
print(result.summary())
# Saves ID mapping: .id_mapping_123456_to_789012.json

Step 2: Agent Modifies Sandbox

The AI agent makes changes in the sandbox using the Rossum API client:

• Modify schemas, hooks, rules
• Test changes with real documents
• Iterate until satisfied

Step 3: Pull Both Workspaces

# Pull original workspace into one directory
source_ws = Workspace("./source-workspace", api_base="...", token="...")
source_ws.pull_workspace(workspace_id=111111)

# Pull modified workspace into another directory
target_ws = Workspace("./target-workspace", api_base="...", token="...")
target_ws.pull_workspace(workspace_id=222222)  # The copied workspace ID

Step 4: Compare Workspaces

# Load the ID mapping from the copy operation
id_mapping = source_ws._load_id_mapping(
    source_org_id=123456,
    target_org_id=789012
)

# Compare source against target
comparison = source_ws.compare_workspaces(
    target_workspace=target_ws,
    id_mapping=id_mapping
)
print(comparison.summary())

Output:

# Workspace Comparison

Source workspace: 111111
Target workspace: 222222

- Identical: 8
- Different: 2
- Source only: 0
- Target only: 1

## Differences

### hook: Invoice Validation (source: 333333 → target: 444444)
  - **config**:
    - Source: {"code": "def validate()..."}
    - Target: {"code": "def validate_v2()..."}

### schema: Invoice Schema (source: 555555 → target: 666666)
  - **content**:
    - Source: [{"id": "invoice_number"...}]
    - Target: [{"id": "invoice_number"}, {"id": "po_number"...}]

## Target Only (not in source)
- rule: New Validation Rule (777777)

Why This Workflow?

• Safe: Changes are made in an isolated sandbox, never touching production
• Transparent: Full diff shows exactly what the agent changed
• Reversible: Original workspace is untouched; discard sandbox if needed
• Auditable: Compare results can be reviewed before deployment

Complete Example: Copy, Modify, Compare, and Push

This example demonstrates the full agent workflow: copy a workspace, make changes in the sandbox, compare against the original, and push changes.

flowchart LR
    subgraph PRD["Production Workspace"]
        P1[Pull PRD]
    end

    subgraph Copy["Copy Operation"]
        C1[copy_workspace]
        C2[ID Mapping]
    end

    subgraph Sandbox["Sandbox Workspace"]
        S1[Pull Sandbox]
        S2[Compare]
    end

    subgraph Agent["AI Agent"]
        A1[Modify via API]
    end

    P1 --> C1
    C1 --> C2
    C2 --> S1
    C2 --> A1
    A1 -->|"Push to Sandbox"| S1
    S1 --> S2
    P1 -.->|"compare_workspaces()"| S2
    S2 -->|"Push to PRD"| P1

import os
from rossum_deploy import Workspace

ROSSUM_API_BASE_URL = "https://api.elis.rossum.ai/v1"
ROSSUM_API_TOKEN = "your-api-token"

# Source workspace to copy
source_workspace_id = 111111
target_org_id = 222222  # Can be same org for testing

# Create separate directories for source and target
source_dir = "./source-workspace"
target_dir = "./target-workspace"

os.makedirs(source_dir, exist_ok=True)
os.makedirs(target_dir, exist_ok=True)

# 1. Pull the source workspace
source_ws = Workspace(source_dir, api_base=ROSSUM_API_BASE_URL, token=ROSSUM_API_TOKEN)
source_ws.pull_workspace(source_workspace_id)
print(f"Pulled source workspace {source_workspace_id}")

# 2. Copy workspace to target org (creates sandbox copy)
copy_result = source_ws.copy_workspace(source_workspace_id, target_org_id)
print(f"Copied to new workspace: {copy_result.id_mapping.mappings['workspace']}")

# 3. Pull the copied target workspace
target_workspace_id = copy_result.id_mapping.mappings["workspace"][source_workspace_id]
target_ws = Workspace(target_dir, api_base=ROSSUM_API_BASE_URL, token=ROSSUM_API_TOKEN)
target_ws.pull_workspace(target_workspace_id)
print(f"Pulled target workspace {target_workspace_id}")

# 4. Make local modifications (e.g., edit a schema field)
# The agent would typically modify files in target_dir here
# For example: change a field's "hidden" property from false to true

# 5. Check local diff (what changed locally vs remote)
local_diff = target_ws.diff()
print(local_diff.summary(color=True))
# Output:
# # Diff Summary
# - Unchanged: 40
# - Local modified: 1
# - Remote modified: 0
# - Conflicts: 0
# - Local only: 0
#
# ## Local Changes (ready to push)
# ### schema: Invoice Schema (4071638)
# #### content
# -        "hidden": false,
# +        "hidden": true,

# 6. Compare source workspace against target (using ID mapping)
comparison = source_ws.compare_workspaces(target_ws, copy_result.id_mapping)
print(comparison.summary(color=True))
# Output:
# # Workspace Comparison
# Source workspace: 111111
# Target workspace: 333333
# - Identical: 30
# - Different: 1
# - Source only: 0
# - Target only: 0
#
# ## Differences
# ### schema: Invoice Schema (source: 123456 → target: 789012)
# #### content
# -        "hidden": false,
# +        "hidden": true,

# 7. Push changes to remote (after review/approval)
push_result = target_ws.push()
print(push_result)
# Output: PushResult(pushed=[(<ObjectType.SCHEMA: 'schema'>, 4071638, 'Invoice Schema')], skipped=[], failed=[])

# 8. Verify everything is in sync
final_diff = target_ws.diff()
print(final_diff.summary())
# Output:
# # Diff Summary
# - Unchanged: 41
# - Local modified: 0
# - Remote modified: 0
# - Conflicts: 0
# - Local only: 0

Key Points:

1. ID Mapping: The copy_result.id_mapping tracks which source objects map to which target objects
2. Two types of diff:
   • ws.diff() - compares local files vs remote API (for push/pull workflow)
   • ws.compare_workspaces() - compares two workspaces using ID mapping (for sandbox comparison)
3. Isolated changes: All modifications happen in the target sandbox; source is never touched
4. Push when ready: After reviewing the comparison, push changes to persist them

Safety

• Use dry_run=True to preview changes before pushing or deploying
• Conflicts detected when both local and remote changed
• Use force=True to override conflicts
• ID mapping automatically handles cross-org references

Git Integration

rossum-deploy automatically uses Git when available to detect conflicts (when both local and remote changed). Without Git, conflict detection is disabled but pull/diff/push still work normally.

See Also

• deployment-manager (PRD2) - Full-featured deployment tool
• rossum-agent - AI agent that uses rossum-deploy
• rossum-mcp - MCP server for Rossum API