DevAll
A zero-code platform for configuring, running, and monitoring multi-agent workflows through a visual console or Python SDK.
Per-dimension scores and reasoning
Evidence shows: regression tests for path traversal vulnerability (tests/test_attachment_upload_filename.py) indicate basic file upload security; README mentions Docker safe run and Git mode, but no least-privilege or user confirmation mechanisms; data flow transparency is limited, only logs and replay mode mentioned; sensitive data handling: .env for API keys, but no encryption or protection details; dependency security: some versions pinned in pyproject.toml, but no vulnerability scanning or audit evidence; external effects: workflows can execute code and publish content, but no approval or restriction mechanisms; rollback not mentioned; source attribution: README lists contributors, but publisher identity unverified. Deductions: lack of user confirmation, rollback, and explicit least-privilege design.
Evidence shows: tests cover memory module and attachment upload, indicating some self-consistency; dependency list complete, but no lock file or version range verification; failure messages: tests catch API errors and return empty lists, but no user-facing error messages. Deductions: dependency availability unverified, failure messages insufficient.
Evidence shows: README describes multiple scenarios (data visualization, 3D generation, deep research) and provides workflow templates; capability boundaries not clearly defined, but extensible via YAML config; trigger precision: workflows triggered by prompts, but no precise trigger conditions; environment fit: supports multiple OS and Docker, but no detailed deployment requirements. Deductions: capability boundaries and trigger precision insufficiently described.
Evidence shows: README provides clear installation and run guides, including Makefile and Docker; information architecture reasonable with directory structure; naming stability: version exists but no changelog; examples and FAQ: tutorials and cases provided; known limitations not explicitly listed; license Apache-2.0; maintenance responsibility: contributor list but no clear maintenance policy. Deductions: missing changelog and known limitations.
Evidence shows: output usability: Python SDK and Web console provided, results obtainable; marginal value: multiple workflow templates save development time; cost-benefit: no performance or cost data. Deductions: cost-benefit lacks data support.
Evidence shows: README cites papers and versions, but no implementation details; cross-source verification: tests exist but not independently verified; fact-inference separation: claims in README not clearly distinguished. Deductions: claims lack traceability, tests not independently verified.
- Publisher identity unverified; exercise caution regarding supply chain risks.
- Dependencies not locked, risk of version drift.
- Lack of user confirmation and rollback mechanisms; workflow execution may have irreversible effects.
- API keys managed via .env; ensure environment security.
What does this agent do, and when should you use it?
ChatDev 2.0, branded DevAll, is a zero-code multi-agent orchestration platform for building and executing configurable workflows. Its FastAPI backend lives in `server/`, while `runtime/` handles agent abstraction and tool execution, `workflow/` handles collaboration logic, and `entity/` supplies configuration-driven orchestration. A Vue 3 web console offers a workflow canvas and Launch view for real-time logs, intermediate artifacts, and human feedback. Workflows are YAML files in `yaml_instance/`, and the Python SDK exposes `runtime.sdk.run_workflow` for programmatic execution and retrieval of the final node message. The repository documents local development and Docker Compose deployment, with example workflows for data visualization, 3D generation, game development, deep research, and teaching videos.
Users configure node parameters and context flow in the Web Console’s Workflow canvas, or select a YAML workflow from yaml_instance/, upload attachments when needed, enter a task prompt, and launch it. The backend executes the workflow; runtime/ manages agent abstraction and tool execution, while the interface exposes real-time logs, intermediate artifacts, and human-in-the-loop feedback. Programmatic callers import run_workflow from runtime.sdk and provide yaml_file, task_prompt, optional attachments, and optional variables; the returned final_message can be read with text_content(). make sync uploads YAML workflows from yaml_instance/ to the database, and make validate-yamls checks every YAML file for syntax and schema errors. Developers can add custom Python tools under functions/.
- A data analyst who needs four to six PNG charts from a large real-estate transactions dataset can run a data-visualization workflow and upload the dataset.
- A team prototyping a Tank Battle game can select
GameDev_v1.yamlorChatDev_v1.yamlin Launch and provide its game request. - A researcher investigating recent advances in LLM-based agent RL can run
deep_research_v1.yamlfor a structured research task. - A creator with Blender and
blender-mcpinstalled can use a 3D-generation workflow for requests such as building a Christmas tree. - A developer automating repeated YAML executions can call
run_workflowwith task prompts, attachments, and per-run variable overrides.
What are this agent's strengths and limitations?
- It combines YAML authoring, a visual workflow canvas, Launch-time execution, log inspection, artifact inspection, and human feedback rather than offering only a code API.
- It supports a local web console, Docker Compose deployment, and
runtime.sdk.run_workflow, covering interactive operation and batch automation. - Its runnable YAML collection includes data visualization, 3D generation, game development, deep research, and teaching-video scenarios.
make validate-yamlschecks workflow syntax and schemas, whilemake syncuploads workflows to the database.
- The development setup requires Python 3.12+, Node.js 18+, and
uv, with separate backend and frontend dependency installation. - Adopters must configure an LLM provider’s
API_KEYandBASE_URL; the README does not identify supported providers, models, or pricing. - The 3D workflows require Blender and
blender-mcp, while the teaching-video workflow requiresuv add manimbefore use. - The documented defaults use ports 6400 and 5173; if 6400 is occupied, both the backend port and
VITE_API_BASE_URLmust be changed together.
How do you install or deploy this agent?
Prerequisites are Python 3.12+, Node.js 18+, and uv. From the repository root, run:
uv sync
cd frontend && npm installThen create configuration:
cp .env.example .envSet API_KEY and BASE_URL in .env for the chosen LLM provider. Optionally, with Docker and Docker Compose installed, run docker compose up --build to start the full application.
How do you use this agent?
After configuring .env, run make dev from the project root and open http://localhost:5173. Alternatively, run uv run python server_main.py --port 6400 --reload, then run VITE_API_BASE_URL=http://localhost:6400 npm run dev from frontend/. In Launch, select a workflow, upload required files, and enter a prompt; use make sync to upload workflow YAML files to the database and make validate-yamls to validate them. A minimal Python invocation is: result = run_workflow(yaml_file="yaml_instance/demo.yaml", task_prompt="Summarize the attached document in one sentence.", variables={"API_KEY": "sk-xxxx"}); then read result.final_message.text_content().
How does this agent compare with similar options?
ChatDev 2.0 (DevAll) is presented as a zero-code multi-agent orchestration platform, while the repository maintains legacy ChatDev 1.x on the chatdev1.0 branch. The legacy version is described as a virtual software company with roles such as CEO, CTO, and Programmer collaborating across design, coding, testing, and documentation; 2.0 broadens the scope to configurable workflows for multiple scenarios.
FAQ
Do I need LLM credentials to run it?
API_KEY and BASE_URL in .env for an LLM provider, and variables passed to run_workflow can override environment values.Is the web console the only interface?
runtime.sdk.run_workflow for executing YAML workflows and retrieving the final node message.Can it run fully offline?
What if the frontend cannot reach the backend?
VITE_API_BASE_URL=http://localhost:6401 for the frontend.How are tool permissions isolated?
runtime/ manages tool execution and that custom Python tools can be added under functions/, but it does not document a permission model, sandboxing, or credential-isolation mechanism.