victor-ai 0.6.0

Open-source agentic AI framework with multi-provider support, workflow orchestration, and extensible tool system

Victor

Open-source agentic AI framework. Build, orchestrate, and evaluate AI agents across 22 providers.

---

Features

┌─────────────────────────────────────────────────────────────┐
│                     VICTOR FRAMEWORK                        │
│                                                             │
│  Agents ─── Teams ─── Workflows ─── Evaluation              │
│    │          │          │              │                    │
│  run()    Sequential   StateGraph    SWE-bench              │
│  stream()  Parallel    YAML DSL      Harnesses              │
│  chat()   Hierarchical Checkpoints   Code Quality           │
│           Pipeline                                          │
│                                                             │
│  24 Providers │ 34 Tool Modules │ 9 Verticals │ 4 Scopes   │
└─────────────────────────────────────────────────────────────┘

• 24 LLM Providers — Cloud (Anthropic, OpenAI, Google, Azure, Bedrock, DeepSeek, Vertex) + local (Ollama, LM Studio, vLLM)
• 34 Tool Modules — File ops, git, shell, web, search, docker, testing, refactoring, analysis
• 9 Domain Verticals — Coding, DevOps, RAG, Data Analysis, Research, Security, IaC, Classification, Benchmark
• Multi-Agent Teams — 4 formations: sequential, parallel, hierarchical, pipeline
• Stateful Workflows — YAML DSL compiled to StateGraph with typed state and checkpointing
• Air-Gapped Mode — Full functionality with local models for secure, offline environments
• Built-in Resilience — Automatic retry with exponential backoff on rate limits, circuit breaker protection

Benchmark Results (March 2026)

Victor achieves 100% task success rate across multiple providers:

Provider	Model	5-Task Success	Avg Time/Task	Cost/1M tokens
Anthropic	Claude Haiku 4.5	100%	16.9s	$0.80 in
OpenAI	GPT-4o-mini	100%	14.7s	$0.15 in
DeepSeek	DeepSeek-Chat V3	100%	35.9s	$0.07 in

Tasks: code generation, research synthesis, file operations, security audit, workflow orchestration. See full results

At a glance

                              ┌─────────────────────────────────┐
                              │       Agent Orchestrator        │
                              │                                 │
[You] ──▶ [CLI/TUI/API] ──▶  │  ProviderManager ──▶ 24 LLMs   │ ──▶ [Response]
                              │  ToolPipeline    ──▶ 34 Tools   │
                              │  TeamCoordinator ──▶ Agents     │
                              │  StateManager    ──▶ 4 Scopes   │
                              └─────────────────────────────────┘

Choose your path

Persona	Start here	Typical goals
New user	Getting Started	Install, first run, local vs cloud
Daily user	User Guide	Commands, modes, profiles, workflows
Operator	Operations	Deployment, monitoring, security
Contributor	Development	Setup, testing, architecture, extending
Architect	Architecture	System overview, core components

Quick start

Path	Commands	Best for
Local model	pipx install victor-ai ollama pull qwen2.5-coder:7b victor chat "Hello"	Privacy, offline, free tier
Cloud model	pipx install victor-ai export ANTHROPIC_API_KEY=... victor chat --provider anthropic	Max capability
Docker	docker pull ghcr.io/vjsingh1984/victor:latest docker run -it -v ~/.victor:/root/.victor ghcr.io/vjsingh1984/victor:latest	Isolated env

Supported Providers

Victor supports 22 LLM providers — switch mid-conversation without losing context.

Category	Providers
Frontier Cloud	Anthropic, OpenAI, Google Gemini, Azure OpenAI
Cloud Platforms	AWS Bedrock, Google Vertex
Specialized	xAI, DeepSeek, Mistral, Groq, Cerebras, Moonshot, ZAI
Aggregators	OpenRouter, Together AI, Fireworks AI, Replicate, Hugging Face
Local (air-gapped)	Ollama, LM Studio, vLLM, llama.cpp

Full Provider Reference

Python API

Victor provides a clean Python API for programmatic use:

from victor.framework import Agent, EventType

# Simple use case
agent = await Agent.create(provider="anthropic")
result = await agent.run("Explain this codebase structure")
print(result.content)

# Streaming responses
async for event in agent.stream("Refactor this function"):
    if event.type == EventType.CONTENT:
        print(event.content, end="")
    elif event.type == EventType.TOOL_CALL:
        print(f"\nUsing tool: {event.tool_name}")

# With tool configuration
from victor.framework import ToolSet

agent = await Agent.create(
    provider="openai",
    model="gpt-4o",
    tools=ToolSet.default()  # or ToolSet.minimal(), ToolSet.full()
)

# Multi-turn conversation
session = agent.chat()
await session.send("What files are in this project?")
await session.send("Now explain the main entry point")

StateGraph Workflows

from victor.framework import StateGraph, END
from typing import TypedDict

class MyState(TypedDict):
    query: str
    result: str

graph = StateGraph(MyState)

graph.add_node("research", research_fn)
graph.add_node("synthesize", synthesize_fn)

graph.add_edge("research", "synthesize")
graph.add_edge("synthesize", END)

compiled = graph.compile()
result = await compiled.invoke({"query": "AI trends 2025"})

Core capabilities

Capability	What it means	Docs
Agent abstractions	run(), stream(), chat(), run_workflow(), run_team()	Framework
22 Providers	Cloud + local LLMs; switch mid-thread without losing context	Providers
33 Tool modules	File ops, git, shell, web, search, docker, testing, analysis	Tool catalog
Workflows	YAML DSL compiled to StateGraph with typed state + checkpointing	Workflows
Multi-agent teams	4 formations: sequential, parallel, hierarchical, pipeline	Multi-agent
State management	4 scopes: workflow, conversation, team, global	State
9 Verticals	Domain-focused agents with tools, prompts, and workflows	Verticals
Evaluation	Agent harnesses, code quality analysis, SWE-bench integration	Evaluation

Command quick reference

Command	Purpose	Example
victor	TUI mode	victor
victor chat	CLI mode	victor chat "refactor this"
victor chat --mode plan	Plan-only analysis	victor chat --mode plan
victor serve	HTTP API	victor serve --port 8080
victor mcp	MCP server	victor mcp --stdio
/provider	Switch provider in chat	/provider openai --model gpt-4

Screenshots

The Victor TUI provides an interactive terminal interface with syntax highlighting and tool status.

CLI mode for quick queries and script integration.

Documentation

• Getting Started
• User Guide
• Guides
• Reference
• Operations
• Development
• Architecture
• Roadmap

Contributing

We welcome contributions. Start with CONTRIBUTING.md and CODE_OF_CONDUCT.md.

Community

• GitHub
• Discussions
• Issues
• Discord

Acknowledgments

Victor is built on the shoulders of excellent open-source projects:

• Pydantic - Data validation and settings management
• Tree-sitter - Incremental parsing for code analysis
• Textual - Modern TUI framework
• Typer - CLI interface with type hints
• Rich - Beautiful terminal formatting
• httpx - Async HTTP client
• Anthropic SDK - Claude API client
• OpenAI SDK - OpenAI API client

License

Apache License 2.0 - see LICENSE.