Neuro SAN Studio: Build Multi-Agent Systems with Declarative Config
A hands-on playground for the Neuro SAN framework to design, test and deploy multi-agent networks in minutes.
- Source repo
- cognizant-ai-lab/neuro-san-studio
- Stars
- ★ 1.2k
- Last updated
- today
- License
- Apache-2.0
- Primary language
- Python
- FA score
- Insufficient evidence
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Researchers exploring adaptive multi-agent systems can experiment with prebuilt examples or custom networks.
- Not a fit if
- Users unwilling to supply paid LLM API keys
- Windows users: install docs cover only Linux and macOS
- Source review
- Insufficient evidence 8 safety controls not found
What does this agent do, and when should you use it?
Neuro SAN Studio is an open-source developer platform from Cognizant AI Lab that serves as a playground for the Neuro SAN multi-agent orchestration framework. It provides ready-to-run examples, tutorials, and tools to design, test, and deploy sophisticated agent networks using declarative HOCON configuration files. The studio includes a command-line tool ns, a web UI called nsflow, and a meta-agent called Agent Network Designer that can generate entire agent networks from natural language descriptions. It supports flexible integration with various LLM providers (OpenAI, Anthropic, Google Gemini) and external tools/frameworks. Installation uses the uv package manager, and initialization is done via ns init. The platform targets researchers, developers, and domain experts, emphasizing data-driven configuration and adaptive communication (AAOSA protocol).
Neuro SAN Studio enables users to build and run multi-agent systems. Specifically, it scaffolds a project structure via ns init, which creates folders like config, mcp, registries, coded_tools, and middleware. Users can import predefined agent networks (e.g., Agent Network Designer and industry examples) with ns import. Running ns run starts the Neuro SAN server (default localhost:8080) and the nsflow UI (localhost:4173). The Agent Network Designer takes natural language input and generates agent network configurations, including agent roles, connections, instructions, and sample queries. The tool also supports ns chat to converse directly with an agent network, ns check-llm-keys to validate API keys, and ns check-config to validate HOCON configurations. These operations ultimately produce interactive multi-agent applications for scenarios like customer support, financial compliance, insurance claims, and more.
- Researchers exploring adaptive multi-agent systems can experiment with prebuilt examples or custom networks.
- Developers can rapidly prototype production solutions, generating configurations from natural language and integrating external APIs.
- Domain experts (e.g., policy admins) can design agent interactions without coding to handle customer inquiries.
- Enterprises can deploy automated customer support, such as airline policy assistance or banking compliance monitoring.
- IT teams can implement internal knowledge management, letting employees query HR and IT policies via agents.
- Developers can use Agent Network Designer to create tailored multi-agent networks for specific industry problems.
How do you install or deploy this agent?
Requires Python and uv (Linux/macOS). In a project folder run:
mkdir my_project && cd my_project
uv init
uv venv
source .venv/bin/activate
uv add neuro-san-studioInitialize the project:
ns initChoose an LLM provider (OpenAI, Anthropic, or Google Gemini) when prompted. Set your API key:
export OPENAI_API_KEY="XXX" # or other provider key, or create .envValidate setup:
ns check-llm-keys
ns check-configHow do you use this agent?
Start the server and UI with ns run, then open http://localhost:4173/ in your browser. Click the NEW button to launch the Agent Network Designer, enter a natural-language description, and it will generate an agent network. Click Launch to test it. Alternatively, import prebuilt networks with ns import basic or ns import hello_world. Use ns chat <agent-name> for direct conversation, and ns export to bundle a network as a .hocon or .zip file. See ns --help for full command details.
What are this agent's strengths and limitations?
- Declarative HOCON configuration enables non-programmers to design agent networks.
- Built-in Agent Network Designer accelerates prototyping by auto-generating networks from NLP descriptions.
- Supports multiple LLM providers, avoiding vendor lock-in and adapting to diverse environments.
- Includes Sly Data for secure handling of sensitive data, not exposing it directly to LLMs.
- Comes with a rich set of industry examples (banking, insurance, telecom) for quick adoption.
- Requires familiarity with
uvand Python; a barrier for those not comfortable with command-line tools. - Mandates LLM API keys, introducing cost and security dependencies on external cloud services.
- HOCON syntax has a learning curve; configuration errors can be hard to debug.
- Official installation instructions omit Windows, complicating adoption for Windows users.
- The framework is relatively new, so community support and third-party integrations are limited.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Neuro SAN Studio: Build Multi-Agent Systems with Declarative Config This agent | Insufficient evidence | CLIFree + model costs | ★ 1.2k | today | Python | — |
| Runner Multi-Agent Terminal | 82 · Some gaps | Desktop appFree | ★ 362 | today | Rust | Codex · Claude Code |
| Vigil Multi-Agent Terminal Orchestrator | 69 · Some gaps | Desktop appFree + model costs | ★ 29 | 15d ago | Swift | Codex · Claude Code |
| WUPHF — A Collaborative Office for AI Employees with a Shared Brain | 66 · Some gaps | CLIFree + model costs | ★ 1.5k | 7d ago | Go | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows a security policy (SECURITY.md) and dependency pinning (requirements.txt), but no concrete implementation or documentation for least privilege, user confirmation, data flow transparency, sensitive data handling, external effects, rollback, or source attribution. Hence all trust criteria scored 0.
Repository contains test files (tests/) and CI workflows, but no clear evidence for self-consistency, dependency availability, or failure messages. Hence all reliability criteria scored 0.
README describes various use cases and audiences, but no detailed explanation of capability boundaries, trigger precision, or environment fit. Hence all adaptability criteria scored 0.
Repository includes README, LICENSE, pyproject.toml, etc., but no clear evidence for information architecture, install notes, naming stability, examples/FAQ, known limitations, versioning/changelog, or maintenance responsibility. Hence all convention criteria scored 0.
Repository describes output usability and marginal value, but no cost-benefit analysis. Hence all effectiveness criteria scored 0.
Repository includes tests and CI, but no evidence for claim traceability, cross-source corroboration, or fact-inference separation. Hence all verifiability criteria scored 0.
- 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.
- Static review cannot verify actual runtime behavior; all scores are inferred from file contents.
- No malicious code or obvious risks found, but permission and data handling details are missing; careful evaluation needed.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Main changes - Neuro SAN Studio now uses neuro-san 0.7.12 and nsflow 0.7.5 (and leaf-common 1.4.6). - Azure OpenAI: since neuro-san 0.7.11, Azure requests use Azure's v1 API with the Responses API and openai_api_version is ignored. It has been removed from the sample config, the .env example and the docs, and each Azure deployment now also names its OpenAI model so its tokens are priced. …View on GitHub ↗ -
Release notes excerpt
Main changes - Neuro SAN Studio now uses neuro-san 0.7.7 and nsflow 0.7.4. Because neuro-san 0.7.7 requires them, pip install -U neuro-san-studio also moves openai to 2.54.0 and langchain-openai to 1.6.2. - Agent networks in subfolders are now exposed as MCP tools with __ in place of /, for example basic__coffee_finder. - nsflow 0.7.4 removes the NSFLOW_PLUGIN_MANUAL_EDITOR and NSFLOW_PLUGIN_EXPORT settings. The manual editor now saves through the Agent Network Designer. - The OpenAI video generation tool is removed, because OpenAI has shut down its Videos API. …View on GitHub ↗ -
Release notes excerpt
What's Changed * ANDe: research named organizations with internet_info_gatherer before designing by @Noravee in https://github.com/cognizant-ai-lab/neuro-san-studio/pull/1397 * Migrate webpage RAG onto SafeFetch with per-URL PDF/HTML routing by @Noravee in https://github.com/cognizant-ai-lab/neuro-san-studio/pull/1366 * Migrate PDF RAG off langchain-community onto SafeFetch and pypdf by @Noravee in https://github.com/cognizant-ai-lab/neuro-san-studio/pull/1400 …View on GitHub ↗
FAQ
What is the cost of using Neuro SAN Studio?
Can it run offline?
How does it handle sensitive data?
How can I debug an agent network?
ns check-config to validate settings.Can I import custom agent networks?
ns import and ns export to share networks as .hocon or .zip files, allowing reuse and collaboration.