binex 0.7.0

Debuggable runtime for AI agent pipelines

Binex

Open-source visual orchestrator for AI agent workflows
Build, run, debug, and replay multi-agent pipelines — 100% locally.

Demo · Install · Web UI · Features · Docs · Issues

---

Demo

1. Start in seconds

_{Install, run binex ui, and you're building workflows}

2. Build & run custom workflows

_{Drag & drop nodes, configure models, run with human input}

3. Explore & debug results

_{Debug, trace, diff, lineage — full post-mortem inspection}

(back to top)

---

What is Binex?

Binex is an open-source, fully local runtime for AI agent workflows. No cloud. No telemetry. No vendor lock-in.

pip install binex
binex ui

That's it. Browser opens. You're building AI workflows.

Why Binex?

• 100% local — your data never leaves your machine
• 100% open source — MIT licensed, audit every line
• Zero telemetry — no tracking, no analytics, no surprises
• Full debuggability — every input, output, prompt, and cost is visible
• Any model — OpenAI, Anthropic, Google, Ollama, OpenRouter, DeepSeek, and 40+ more via LiteLLM

(back to top)

---

Installation

Requires Python 3.11+

pip install binex

With extras:

pip install binex[langchain]    # LangChain Runnables
pip install binex[crewai]       # CrewAI Crews
pip install binex[autogen]      # AutoGen Teams
pip install binex[telemetry]    # OpenTelemetry tracing
pip install binex[rich]         # Rich colored CLI output

(back to top)

---

Web UI

Launch the visual workflow editor:

binex ui

Visual Drag & Drop Editor

_{Collapsible node sections, tool picker with 10 built-in tools, MCP config, Visual ↔ YAML sync}

6 node types: LLM Agent, Local Script, Human Input, Human Approve, Human Output, A2A Agent

• 20+ preset models including 8 free OpenRouter models
• Built-in prompt library (Planner, Researcher, Analyzer, Writer, Reviewer, Summarizer)
• Tool Picker — 10 built-in tools, MCP server integration, custom Python tools
• Collapsible sections — Model, Prompt, Tools, Advanced per LLM node
• Workflow Settings panel — configure MCP servers (stdio/HTTP) and cron schedules
• Switch between Visual and YAML modes — changes sync both ways (including tools & MCP)
• Real-time cost estimation as you build
• Custom model input — use any litellm-compatible model

Dashboard

_{All runs at a glance — status, cost, duration}

Debugging & Analysis

_{Left: Node-by-node debug inspection. Right: Gantt timeline with anomaly detection.}

Run Comparison

_{Side-by-side diff with filtering: changed, failed, cost delta}

19 Pages — Full CLI Parity

Category	Pages
Workflows	Browse, Visual Editor (with tool picker & MCP config), Scaffold Wizard
Runs	Dashboard, RunLive (SSE), RunDetail
Analysis	Debug (input/output artifacts), Trace (Gantt timeline), Diagnose (root-cause), Lineage (artifact graph)
Comparison	Diff (side-by-side with filter bar, compare with previous run), Bisect (NodeMap, DAG visualization, divergence metrics)
Costs	Cost Dashboard (charts), Budget Management
System	Doctor (health), Plugins, Gateway, Export, Scheduler (cron)

Navigation

Sidebar organized into 4 groups: Build (Editor, Scaffold), Runs (Dashboard), Analyze (Compare, Bisect), System (Gateway, Plugins, Doctor). Run-specific pages (Debug, Trace, Diagnose, Lineage, Costs) open from run context.

Replay

Debug any node → click Replay → swap the model or prompt → re-run just that node. No re-running the entire pipeline.

(back to top)

---

Quickstart

CLI

# Zero-config demo
binex hello

Tip: Runs a 2-node demo workflow (producer → consumer), no API keys needed.

# Run a workflow
binex run examples/simple.yaml

Tip: Uses your configured LLM provider. Set OPENAI_API_KEY or use ollama for fully local runs.

# Inspect the run
binex debug latest
binex trace latest

Tip: debug shows per-node inputs/outputs. trace shows the execution timeline as a Gantt chart.

Web UI

binex ui

Tip: Opens the browser automatically. Use --port 9000 to change the port, --no-browser to skip auto-open.

Create a Workflow

name: research-pipeline
nodes:
  input:
    agent: "human://input"
    outputs: [output]

  planner:
    agent: "llm://gemini/gemini-2.5-flash"
    system_prompt: "Break this topic into research questions"
    tools:
      - "builtin://web_search"
      - "builtin://calculator"
    depends_on: [input]
    outputs: [output]

  researcher:
    agent: "llm://openrouter/google/gemma-3-27b-it:free"
    system_prompt: "Investigate and report findings"
    tools:
      - "builtin://fetch_url"
    depends_on: [planner]
    outputs: [output]

  output:
    agent: "human://output"
    depends_on: [researcher]
    outputs: [output]

(back to top)

---

Features

Agent Adapters

Prefix	Description
local://	In-process Python callable
llm://	LLM via LiteLLM (40+ providers)
a2a://	Remote agent via A2A protocol
human://input	Free-text input from user
human://approve	Approval gate with conditional branching
human://output	Display results to user
builtin://	10 built-in tools (calculator, web_search, shell_command, etc.)
mcp://	MCP server tools (stdio or HTTP transport)
python://	Custom Python function as tool
langchain://	LangChain Runnable (plugin)
crewai://	CrewAI Crew (plugin)
autogen://	AutoGen Team (plugin)

CLI Commands

Command	Description
binex run	Execute a workflow
binex ui	Launch Web UI
binex debug	Post-mortem inspection
binex trace	Execution timeline
binex replay	Re-run with agent swaps
binex diff	Compare two runs
binex diagnose	Root-cause failure analysis
binex bisect	Find first divergence between two runs
binex cost show	Cost breakdown per node
binex explore	Interactive TUI dashboard
binex scaffold	Generate workflow from DSL
binex export	Export to CSV/JSON
binex doctor	System health check
binex hello	Zero-config demo
binex list	List available workflows
binex start	Create a new project interactively
binex init	Deprecated alias for binex start
binex validate	Validate workflow YAML
binex cancel	Cancel a running workflow
binex artifacts	Inspect artifacts
binex dev	Local development environment
binex gateway	A2A Gateway management
binex plugins	Manage adapter plugins
binex workflow	Workflow versioning & inspection
binex scheduler start	Start cron-based workflow scheduler
binex scheduler list	List scheduled workflows
binex scheduler add/remove	Register/unregister workflow files

LLM Providers

OpenAI · Anthropic · Google Gemini · Ollama · OpenRouter · Groq · Mistral · DeepSeek · Together AI

Built With

(back to top)

---

Examples

29 example workflows in examples/. Highlights:

Example	What it demonstrates
simple.yaml	Minimal two-node pipeline
diamond.yaml	Diamond dependency pattern
fan-out-fan-in.yaml	Parallel execution with aggregation
human-in-the-loop.yaml	Approval gates and conditional branching
human-feedback.yaml	Human feedback loop
conditional-routing.yaml	Conditional node execution
multi-provider-demo.yaml	Multiple LLM providers in one workflow
ollama-research.yaml	Full research pipeline with Ollama + OpenRouter
budget-hard-limit.yaml	Budget enforcement with hard stop
budget-per-node.yaml	Per-node budget allocation
a2a-multi-agent.yaml	A2A protocol multi-agent workflow
langchain-summarizer.yaml	LangChain Runnable in a pipeline
crewai-research-crew.yaml	CrewAI Crew as a workflow node
autogen-coding-team.yaml	AutoGen Team for code generation
mixed-framework-pipeline.yaml	LangChain + CrewAI + AutoGen in one pipeline

(back to top)

---

Architecture

src/binex/
├── adapters/        # Agent backends (local, LLM, A2A, human, frameworks)
├── agents/          # Agent definitions and configuration
├── cli/             # Click CLI commands
├── gateway/         # A2A Gateway server and routing
├── graph/           # DAG construction + topological scheduling
├── models/          # Pydantic v2 domain models
├── plugins/         # Plugin registry for custom adapters
├── prompts/         # 112 built-in prompt templates
├── registry/        # Provider and adapter registry
├── runtime/         # Orchestrator, dispatcher, replay engine
├── scheduler/       # Cron-based workflow scheduling
├── stores/          # SQLite execution + filesystem artifacts
├── tools/           # @tool decorator, 10 built-in tools, MCP client
├── trace/           # Debug, lineage, timeline, diffing
├── ui/              # FastAPI backend + React frontend
│   ├── api/         # 32 REST endpoints
│   └── static/      # Pre-built React app
└── workflow_spec/   # YAML loader + validator

(back to top)

---

Documentation

Full docs at alexli18.github.io/binex

(back to top)

---

Contributing

Contributions are welcome! If you find this useful:

• Star the repo — it takes 1 second and helps more than you know
• Open issues — tell me what's broken or what you need
• Submit PRs — let's build this together

I'm a solo developer building this in the open. Every star, issue, and PR makes a real difference.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

(back to top)

---

Configuration

Binex can be configured via environment variables:

Variable	Default	Description
BINEX_STORE_PATH	.binex	Directory for SQLite database and artifacts
BINEX_DEFAULT_DEADLINE_MS	120000	Default node timeout in milliseconds
BINEX_DEFAULT_MAX_RETRIES	1	Default retry count for failed nodes
BINEX_DEFAULT_BACKOFF	exponential	Retry backoff strategy (fixed or exponential)
BINEX_REGISTRY_URL	http://localhost:8000	Default A2A agent registry URL

All data is stored in .binex/ (gitignored by default):

• .binex/binex.db — SQLite database (runs, execution records, cost records)
• .binex/artifacts/ — JSON artifact files

(back to top)

---

License

Distributed under the MIT License. See LICENSE for more information.

(back to top)

---

_{Built by a solo dev who believes AI agents shouldn't be black boxes.
No cloud. No telemetry. No surprises. Just debuggable AI workflows.}