ProAgent Workflow Automation

Turns human instructions into n8n workflows and coordinates specialized agents for workflow decisions.

Source repo
OpenBMB/ProAgent
Stars
★ 865
Last updated
2y ago
License
Apache-2.0
Primary language
Python

At a glance

Works with
Platform-specificOpenAI API (Partial support)
You'll need
Pythonn8nOpenAI APIShell / CLINetwork accessLocal filesystem
Typical use
An operations team with self-hosted n8n that wants to construct cross-application workflows from natural-language requests.
Main limitation
Core operation depends on OpenAI configuration and is explicitly based on GPT4-0613 plus an older OpenAI interface version, creating version-compatibility risk.

What does this agent do, and when should you use it?

ProAgent is the official code for the Agentic Process Automation (APA) paper, intended to turn human instructions into automation workflows. It operates around self-hosted n8n and expects n8n-exported workflow and credential files to be supplied to the codebase. Runtime behavior is configured in ProAgent/config.py: development builds workflows as described in the paper, refine modifies an existing workflow, and production loads an existing run for reproduction. Each run creates a readable record under ./records that refine and production can load later. Development requires OpenAI API configuration; the repository states that the implementation is based on GPT4-0613 and an older OpenAI interface version.

After a runtime mode is selected in ProAgent/config.py, the entry point is python main.py. In development mode, ProAgent builds workflows from human instructions, coordinates specialized agents for complex decisions, and can ask a human for help through a function call; n8n must be started first. The code consumes c.json and w.json exported through n8n export:credentials and n8n export:workflow, then writes run records to ./records. Refine loads an existing workflow and changes it for a new request with test-on-change enabled; production loads an existing run to reproduce it and tests APA-code only once at the end.

  1. An operations team with self-hosted n8n that wants to construct cross-application workflows from natural-language requests.
  2. A researcher reproducing the reported paper cases by loading an existing run from ./apa_case in production mode.
  3. An automation maintainer who has an existing n8n workflow and needs to refine it for a new request.
  4. A team building and testing workflows with human intervention when the system needs help understanding a request.
  5. A developer who has registered applications and credentials in n8n and needs the code to load workflow and credential IDs.

How do you install or deploy this agent?

Install Python packages:

pip install -r requirements.txt

To connect to real application services, install and run self-hosted n8n:

npm install n8n -g
export WEBHOOK_URL=https://n8n.x-agent.net/redirect/http%3A%2F%2Flocalhost%3A5678/
n8n

After connecting or registering apps in n8n, export credentials and workflows:

n8n export:credentials --all --decrypted --output=./ProAgent/n8n_tester/credentials/c.json
n8n export:workflow --all --output=./ProAgent/n8n_tester/credentials/w.json

Development mode additionally requires OPENAI_API_KEY and OPENAI_API_BASE.

How do you use this agent?

Select development, refine, or production in ProAgent/config.py. For development, start n8n and set OPENAI_API_KEY plus OPENAI_API_BASE, then run:

python main.py

To reproduce a saved case, use production mode to load an existing run from ./apa_case or ./records; this reproduction path does not require n8n. Use refine mode to load an existing workflow and improve it for a new request.

What are this agent's strengths and limitations?

Pros
  • Combines workflow construction and execution-time decisions in one APA process instead of only running pre-authored RPA steps.
  • Provides distinct development, refine, and production modes for construction, iterative changes, and run reproduction.
  • Writes readable records under ./records that can be reloaded by refine or production.
  • Includes a function-call path for proactively requesting human help when the request has been misunderstood.
Limitations
  • Core operation depends on OpenAI configuration and is explicitly based on GPT4-0613 plus an older OpenAI interface version, creating version-compatibility risk.
  • Connecting real applications requires self-hosted n8n, pre-registered apps, and exported decrypted credentials and workflows.
  • The repository states that its n8n compiler targets an older n8n version and may not work with newer releases.
  • Self-hosted n8n setup and third-party account connections can encounter unaddressed problems; the repository also notes possible network restrictions in China.

How does this agent compare with similar options?

Against traditional RPA, ProAgent is positioned for workflow design and dynamic execution decisions that need human-like intelligence. Rather than only following fixed automation steps, it attempts to construct workflows from human instructions and coordinate specialized agents for decisions.

Key facts side by side with the most closely related agents.

Agent Source review Stars Updated Language Full support on
ProAgent Workflow Automation This agent 28 · Major gaps ★ 865 2y ago Python
superglue Enterprise Integration Agents 36 · Major gaps ★ 2.1k 1mo ago TypeScript OpenAI API · Claude API
LocalAI – Open-Source AI Engine 53 · Major gaps ★ 49k 4d ago Go OpenAI API · Claude API
Hacker Podcast – AI-generated Chinese Hacker News Daily 40 · Major gaps ★ 2.6k 21d ago TypeScript OpenAI API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
28/ 100 5-point scale 1.4 / 5
Trust 0/29
Reliability 3/14
Adaptability 8/18
Convention 8/18
Effectiveness 6/13
Verifiability 3/8
Why each dimension lost points
Trust0 / 29 · 0.0/5

Evidence: The repository provides no explicit mechanisms for least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, or source attribution. README mentions using n8n and OpenAI API but does not describe permission minimization or user confirmation. Therefore all trust criteria score 0.

Reliability3 / 14 · 1.1/5

Evidence: README describes three running modes (development, refine, production) but lacks details on error handling or failure messages. Dependencies are pinned but availability is not discussed. Thus self_consistency scores 1, dependency_availability scores 1, failure_messages scores 0.

Adaptability8 / 18 · 2.2/5

Evidence: README identifies target audience (researchers and developers) and use cases (APA workflow construction). However, capability boundaries are unclear, trigger conditions (e.g., user instructions) are not detailed, and environment fit (e.g., n8n version compatibility) is mentioned but limited. Therefore audience_and_scenarios scores 2, capability_boundaries scores 1, trigger_precision scores 1, environment_fit scores 1.

Convention8 / 18 · 2.2/5

Evidence: README provides installation and running instructions but lacks FAQ, changelog, and versioning. License is Apache-2.0, but maintenance responsibility is unclear. Thus information_architecture scores 1, install_notes scores 2, naming_stability scores 1, examples_and_faq scores 1, known_limitations scores 2, license scores 3, versioning_changelog scores 0, maintenance_responsibility scores 1.

Effectiveness6 / 13 · 2.3/5

Evidence: README claims ProAgent can construct and execute workflows but does not provide output format or quality assessment. Marginal value is described (automating complex tasks), but cost-benefit is not quantified. Therefore output_usability scores 1, marginal_value scores 2, cost_benefit scores 1.

Verifiability3 / 8 · 1.9/5

Evidence: README cites a paper but does not provide reproducible experimental details. No independent sources verify claims. Facts and inferences are mixed. Thus claim_traceability scores 1, cross_source_corroboration scores 1, fact_inference_separation scores 1.

Risks and how to mitigate them
  • Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
  • Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
  • Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
  • Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
  • The repository lacks permission management or user confirmation mechanisms; assess risks before use.
  • It depends on n8n and OpenAI, but data flow and sensitive information handling are not described; proceed with caution.
  • README mentions using an external redirect service that may be unstable and is not open-sourced.
Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision 02a79ac62e7a
See the full review method →

FAQ

Does it require OpenAI?
Development mode requires OPENAI_API_KEY and OPENAI_API_BASE. The repository identifies GPT4-0613 and an older OpenAI interface version, and does not document support for other model providers.
Is n8n always required?
Development for real application services requires a running self-hosted n8n instance. To reproduce a supplied recorded case, production mode can load an existing run without preparing n8n.
How does it obtain access to third-party apps?
Apps must first be connected or registered in n8n. Credentials are exported in decrypted form to ./ProAgent/n8n_tester/credentials/c.json, alongside the exported workflow file.
Will it work unchanged with current n8n releases?
That is not guaranteed. The repository says its n8n compiler was built against an older n8n version and may be incompatible with newer versions.
View on GitHub ↗ Install ↓

Related agents