Bond

The Forensic Runtime for AI Agents

See exactly what your agent is thinking. Watch tool calls form in real-time. Observe every execution and result. Full transparency—nothing hidden.

pip install bond-agent

Get Started    API Reference

• Full-Spectrum Streaming

  ---

  See every thought, tool call, and result as it happens. on_thinking_delta, on_tool_call_delta, on_tool_result—nothing is hidden.

  Architecture

• Dynamic Personas

  ---

  Switch agent personas mid-conversation without creating new instances. The same agent, different cognitive frames.

  Quickstart

• Type-Safe Tools

  ---

  Built on PydanticAI. All tool inputs and outputs are validated against schemas—no hallucinated arguments.

  API Reference

• Production Ready

  ---

  WebSocket, SSE, and print handlers included. Plug Bond into any UI framework with minimal glue code.

  Architecture

---

Quick Example

from bond import BondAgent, StreamHandlers

# Create handlers for transparency
handlers = StreamHandlers(
    on_thinking_delta=lambda t: print(f"[Thinking] {t}", end=""),
    on_tool_call_delta=lambda name, args: print(f"[Tool] {name}: {args}"),
    on_tool_result=lambda id, name, res: print(f"[Result] {name}: {res}"),
    on_complete=lambda data: print(f"\n[Done] {data}"),
)

# Create agent with streaming
agent = BondAgent(
    name="forensic-assistant",
    instructions="You are a helpful assistant. Think step by step.",
    model="anthropic:claude-sonnet-4-20250514",
)

# Run with full observability
result = await agent.run(
    "What's 2 + 2?",
    handlers=handlers,
)

---

Why Bond?

Feature	Raw API	LangChain	Bond
Streaming	Text only	Callback-based	Full-spectrum events
Thinking visibility	Hidden	Partial	Real-time deltas
Tool call streaming	After completion	After completion	As it forms
Persona switching	New client	New chain	Same instance
Type safety	Manual	Optional	Built-in (PydanticAI)

Bond's philosophy: The agent's reasoning should be as observable as its output.

---

StreamHandlers: The Forensic Core

@dataclass
class StreamHandlers:
    # Lifecycle: Block open/close
    on_block_start: Callable[[str, int], None] | None = None
    on_block_end: Callable[[str, int], None] | None = None

    # Content: Incremental deltas
    on_text_delta: Callable[[str], None] | None = None
    on_thinking_delta: Callable[[str], None] | None = None
    on_tool_call_delta: Callable[[str, str], None] | None = None

    # Execution: Tool running/results
    on_tool_execute: Callable[[str, str, dict], None] | None = None
    on_tool_result: Callable[[str, str, str], None] | None = None

    # Lifecycle: Response complete
    on_complete: Callable[[Any], None] | None = None

Eight callbacks. Complete observability. Zero black boxes.

---

Get Started

Quickstart Guide