bentotruck 0.1.0

An agent engineering framework — every compartment of the bento box is a building block for agentic systems.

bentotruck

BentoTruck is an agent engineering framework — every compartment of the bento box is a building block for agentic systems.

Just like a bento box: each compartment holds a distinct, purposeful piece — together they make a complete meal.

Philosophy

BentoTruck is not "yet another AI framework." It is Agent Engineering — focused on the fundamental pieces that every agentic system needs, regardless of which model or provider sits underneath.

Not focused on prompts. Not focused on OpenAI. Not focused on LangChain. Focused on architecture.

Compartments

Module	Emoji	Role
rice	rice	Core Agent — Agent, Session, Context, State, Events
teriyaki	meat_on_bone	Planning — ReAct, Tree, Graph, Step, Goal planners
edamame	broccoli	Memory — Conversation, Vector, Graph, Working, LongTerm
gyoza	dumpling	Tools — Python, REST, SQL, Filesystem, Browser, MCP, Docker
tempura	fried_shrimp	Skills — Reusable behaviors: Search, Summarize, Code, Research
nigiri	sushi	Models — Vendor-neutral providers: OpenAI, Claude, Gemini, Ollama, Azure, vLLM
miso	bowl_with_spoon	Reflection — Reflect, Retry, Improve, Critique, Verify
yuzu	tangerine	Evaluation — Confidence, Grounding, Latency, Hallucination, Cost, Quality
onigiri	rice_ball	Workflows — Workflow, Step, Condition, Loop
bento	bento_box	Multi-Agent — Team orchestration
sake	sake	Communication — Message, Broadcast, Channel, Event, Topic
udon	steaming_bowl	Pipelines — Pipeline, Step, Condition, Loop, Parallel
dango	dango	Prompts — Template, SystemPrompt, FewShot, Variables, PromptRepository
naruto	fish_cake	Routing — Capability, Semantic, Rule, Intent, Hybrid routers
katsu	curry	Safety — PII, Injection, Moderation, OutputValidator, PolicyEngine
wasabi	herb	Observability — Trace, Span, Logger, Metrics, Profiler

Quick Example

from bentotruck import rice, naruto, gyoza, edamame, nigiri

agent = (
    rice.Agent("Restaurant Assistant")
    .using(nigiri.Claude())
    .with_memory(edamame.VectorMemory())
    .with_router(naruto.SemanticRouter())
    .with_tools(
        gyoza.SQLTool(),
        gyoza.RESTTool(),
        gyoza.PythonTool(),
    )
)

Fleet

• ThaiTruck — batch DataFrame cleaning & processing (Data Engineering)
• SushiTruck — streaming ingestion & API connectors (Data Acquisition)
• RamenTruck — ML/AI toolkit (Machine Learning)
• BentoTruck — agent engineering framework (Agent Engineering) (this package, in development)

---

Package name secured. Full release coming soon.