AgentOpsCLI

AgentOpsCLI is a commercial-grade, production-oriented AI coding agent and terminal operations platform implemented entirely in Python.

It provides an agentic pair-programming loop with native local Ollama support, OpenAI-compatible APIs, Anthropic Messages API, transparent multi-provider routing, intelligent context discovery, and safe agent tools.

Key Features

Unified Multi-Provider Support: Seamlessly interact with local models via Ollama (llama3.1:8b, qwen2.5-coder:14b), OpenAI (gpt-4o, o3-mini), Anthropic (claude-3-5-sonnet), and Google Gemini.
Intelligent Model Router: Real-time scoring by observed latency, cost, and health, with automatic circuit breaking and bounded fallback.
Local-Only Privacy Mode: Strict privacy enforcement (LOCAL_ONLY=true) guarantees that no code or project data ever leaves your machine.
Project-Aware Context Engine: Respects .gitignore and default exclusions (.git, node_modules, __pycache__, .venv, binaries, and large files), ranking files by relevance to fit strict token budgets.
Safe Agent Tools:
- read_file: Workspace containment and directory traversal protection.
- edit_file: Generates colored unified diff previews and enforces explicit user confirmation before writing changes atomically.
- run_command: Enforces approval before execution, applies strict timeouts, and automatically masks credentials and secrets in output.
Session & Cost Management: Persistent atomic storage in ~/.agentops/sessions, resuming past conversations, and tracking input/output tokens and estimated costs.
Rich Terminal Interface: Syntax-highlighted code blocks and diffs, real-time streaming, and interactive slash commands (/help, /model, /provider, /sessions, /cost, /doctor, /clear).

Architecture Overview

agentops/
├── cli/
│   ├── main.py              # Command-line entrypoint & argument parser
│   └── commands/            # doctor, profile, mcp, auth, tasks, sessions
├── core/
│   ├── agent.py             # AgentOpsEngine main agentic execution loop
│   ├── conversation.py      # Conversation turns and system prompts
│   ├── events.py            # Event emitter and streaming lifecycle
│   └── errors.py            # Typed error hierarchy
├── providers/
│   ├── base.py              # BaseProvider interface & data models
│   ├── ollama.py            # Native Ollama client (streaming & local inference)
│   ├── openai_compatible.py # OpenAI / DeepSeek / vLLM adapter
│   ├── anthropic.py         # Anthropic Messages API adapter
│   ├── gemini.py            # Google Gemini adapter
│   └── registry.py          # Provider lifecycle registry
├── router/
│   ├── router.py            # SmartRouter transparent routing engine
│   ├── scoring.py           # Latency, cost, and reliability scoring
│   ├── policies.py          # Task complexity classification & privacy rules
│   └── fallback.py          # Circuit breaking and bounded retries
├── context/
│   ├── indexer.py           # Workspace file discovery and filtering
│   ├── ignore.py            # Gitignore pattern matching
│   ├── token_budget.py      # Token limits and safe truncation
│   └── builder.py           # Relevance ranking & delimited context assembler
├── tools/
│   ├── base.py              # BaseTool interface & JSON schema generation
│   ├── read_file.py         # Safe file reader
│   ├── edit_file.py         # Diff preview & atomic file editor
│   ├── run_command.py       # Sandboxed command execution & secret masking
│   ├── diff.py              # Unified diff generation & ANSI formatting
│   └── registry.py          # Tool discovery and execution registry
├── sessions/
│   ├── store.py             # Atomic session persistence
│   ├── history.py           # Session & Turn models
│   └── costs.py             # Token & USD cost tracking
├── security/
│   ├── paths.py             # Path boundary validation & traversal protection
│   ├── approvals.py         # Explicit human approval enforcement
│   └── secrets.py           # Secret & credential redaction
└── ui/
    ├── terminal.py          # Rich terminal renderer
    ├── repl.py              # Interactive prompt loop
    ├── themes.py            # Dark & Light palettes
    └── shortcuts.py         # Slash command handlers

Requirements & Installation

Requirements

Python 3.10, 3.11, or 3.12
Git

Installation

Install via pip in editable development mode:

git clone https://github.com/mohdshamii/AgentOps-CLI.git
cd AgentOps-CLI
pip install -r requirements-dev.txt
pip install -e .

Verify the installation:

agentops --version
agentops doctor

The CLI executable is registered as agentops, with backward-compatible aliases mycodecli and agentopscli.

Quickstart & Daily Use

1. Interactive Session (Default)

Start the interactive REPL:

agentops

Inside the REPL, type queries or use slash commands:

/help — Display shortcuts and commands
/model <name> — Inspect or switch the active model
/provider <name> — Inspect or switch the active provider
/sessions — List saved conversations
/cost — View token usage and estimated cost
/doctor — Run runtime health diagnostics
/clear — Clear the terminal screen
/exit — Exit safely

2. Print / Pipe Mode (Non-interactive)

agentops -p "Explain the router architecture in agentops/router/router.py"

Output formatted as JSON for CI or automation:

agentops -p "List all functions in calculator.py" --output-format json

3. Provider Setup & Profiles

Configure a profile quickly:

# Set up for local Ollama
agentops profile init --provider ollama --model llama3.1:8b --goal coding

# Auto-detect best provider
agentops profile auto

Configuration Reference

Configuration can be set via .env, environment variables, .agentops-profile.json, or agentops.yaml.

Variable	Description	Default
`PREFERRED_PROVIDER`	Default provider (`ollama`, `openai`, `anthropic`, `gemini`)	`ollama`
`ROUTER_MODE`	Router mode (`smart`, `fixed`)	`smart`
`ROUTER_STRATEGY`	Routing strategy (`balanced`, `latency`, `cost`)	`balanced`
`ROUTER_FALLBACK`	Automatically fall back on error (`true`, `false`)	`true`
`LOCAL_ONLY`	Enforce strict local-only privacy (`true`, `false`)	`false`
`OLLAMA_BASE_URL`	Ollama service endpoint	`http://localhost:11434`
`OPENAI_API_KEY`	OpenAI API key	None
`ANTHROPIC_API_KEY`	Anthropic Claude API key	None
`GEMINI_API_KEY`	Google Gemini API key	None

Security & Privacy Guarantees

Path Boundary Containment: Every file operation strictly validates that target paths resolve within the active workspace root. Path traversal attempts (../../) raise immediate security violations.
Explicit Approval Workflow:
- File edits show a unified diff and require user confirmation before applying.
- Terminal commands show the exact command and working directory and require explicit approval.
- The agent cannot approve its own actions.
Secret Redaction: API keys and tokens matching credential patterns are masked automatically in command output and logs.

Testing

Run the test suite:

pytest -v

All 80 tests cover unit components and end-to-end integration workflows without external network dependencies.

License

This project incorporates third-party open-source components with verified MIT and BSD-family licenses. For full details, see THIRD_PARTY_LICENSES.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support