CLI agent for the Bob Agent Framework.

bob-cli is a command-line interface for Bob, a general-purpose AI agent framework. It provides:

• Interactive REPL: Chat through AgentLoop with built-in slash commands and session reset shortcuts
• Multi-Model Support: Works with OpenAI, Anthropic, Google, and other LLM providers
• Tool Integration: Connect to MCP servers for file operations, shell commands, and more
• Skill System: Load and apply predefined skills for specialized tasks

# Clone the repository
git clone https://github.com/longcipher/bob
cd bob

# Build and run
cargo run --bin bob-cli -- --config agent.toml

# Download from GitHub releases
# (Coming soon)

Create an agent.toml file in the project root:

[runtime]
default_model = "openai:gpt-4o-mini"
max_steps = 12
turn_timeout_ms = 90000
dispatch_mode = "native_preferred"

# Optional: Configure MCP servers
[mcp]
[[mcp.servers]]
id = "filesystem"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
tool_timeout_ms = 15000

# Optional: Configure skills
[skills]
max_selected = 3
token_budget_ratio = 0.1

[[skills.sources]]
type = "directory"
path = "./skills"
recursive = false

# Optional: Persist session state/checkpoints/artifacts across restarts
[store]
path = "./.bob/sessions"

# Optional: Configure policies
[policy]
deny_tools = ["local/shell_exec"]
allow_tools = ["local/read_file", "local/write_file"]
default_deny = false

# Optional: Configure approval guardrails
[approval]
mode = "allow_all"
deny_tools = ["local/shell_exec"]

# Optional: Configure per-session token budget
[cost]
session_token_budget = 10000

When [store] is set, budget accounting is also persisted, so restarting the CLI does not reset per-session token limits.

dispatch_mode supports native_preferred and prompt_guided.

Set your LLM provider API key:

# For OpenAI
export OPENAI_API_KEY="sk-..."

# For Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."

# For Google
export GEMINI_API_KEY="..."

cargo run --bin bob-cli

The agent will start an interactive REPL:

Bob agent ready  (model: openai:gpt-4o-mini)
Type a message and press Enter. /help for commands.

> Summarize the latest meeting notes in this folder

• Type your message and press Enter to send
• /quit or /exit to exit the agent
• /help or /h to show available commands
• /usage to inspect cumulative session token usage
• /tools to list all available tools
• /tool.describe <tool-name> to print a tool schema
• /tape.info, /tape.search <query>, /anchors, and /handoff [name] for session tape inspection
• /new or /reset to start a fresh session context

> Read docs/design.md and explain it in simple terms

I'll read docs/design.md for you...

[uses filesystem tool to read the document]

The design document describes...

> Translate this explanation into Chinese

[agent produces translated output]

Here is the translated explanation...

Works with any LLM provider supported by liter-llm:

• OpenAI: openai:gpt-4o, openai:gpt-4o-mini
• Anthropic: anthropic:claude-3-5-sonnet-20241022
• Google: google:gemini-2.0-flash-exp
• Groq: groq:llama-3.3-70b-versatile
• And more...

Connect to MCP servers for extended capabilities:

• Filesystem: Read, write, and manage files
• Shell: Execute shell commands
• Database: Query databases
• Custom: Build your own MCP servers

Apply predefined skills for specialized tasks:

• Knowledge extraction
• Summarization and transformation
• Workflow automation
• Domain-specific orchestration

Sessions are persisted in-memory (development) or can be configured for persistent storage.

# Run in development mode
cargo run --bin bob-cli

# Build release binary
cargo build --bin bob-cli --release

# Run tests
cargo test -p cli-agent

The CLI agent is the composition root that:

1. Loads configuration from agent.toml
2. Wires up adapters (LLM, tools, storage, events)
3. Creates the runtime
4. Runs the REPL loop

See the main.rs for the implementation.

• bob-core - Domain types and ports
• bob-runtime - Runtime orchestration
• bob-adapters - Adapter implementations

Licensed under the Apache License, Version 2.0. See LICENSE for details.