Flyto2 Pro Core

Open-source foundation for Flyto2 commercial-grade automation: workflow contract validation, deterministic agent runtime, budget and token metering, provider interfaces, project state, and evidence-backed verification.

Use Flyto2 Pro Core when you need bring-your-own LLM providers, Qdrant/OpenAI integrations, budget enforcement, deterministic workflow validation, or a clean boundary between Apache-2.0 runtime primitives and private enterprise modules.

Official links: flyto2.com · Docs · Flyto2 Core · Flyto2 Indexer · Security

What's Inside

Install

pip install flyto-pro-core

With optional providers:

pip install flyto-pro-core[openai]    # OpenAI LLM + embeddings
pip install flyto-pro-core[qdrant]    # Qdrant vector store
pip install flyto-pro-core[core]      # Flyto2 Core module catalog
pip install flyto-pro-core[factory]   # Blueprint composition + OpenAI fallback
pip install flyto-pro-core[full]      # All providers

Usage

Contract Engine — Validate Workflows

from flyto_pro_core.contract.engine import ContractEngine

engine = ContractEngine()
await engine.initialize()  # loads module catalog from flyto-core

report = await engine.validate_workflow(spec)
if not report.valid:
    for issue in report.issues:
        print(f"  {issue.severity}: {issue.message}")

# Binding resolution
bindings = await engine.get_available_bindings(spec, "node_3")

# Compile to execution plan
plan = await engine.compile(spec)

Cost Controller — Budget Management

from flyto_pro_core.cost.controller import CostController, BudgetConfig

# Per-tier budgets
controller = CostController(budget=BudgetConfig.for_tier("pro"))

# Record usage
controller.record_llm_usage("gpt-4o", prompt_tokens=1000, completion_tokens=500)
controller.record_tool_call()

# Check budget (raises BudgetExceededError if over)
controller.check_budget()

# Summary
print(controller.get_summary())
# {"cost_spent_usd": 0.025, "cost_budget_usd": 1.0, "tokens_used": 1500, ...}

Interfaces — Bring Your Own Provider

from flyto_pro_core.interfaces.llm import ILLMService, LLMResponse
from flyto_pro_core.interfaces.storage import IVectorStoreRepository

# Use built-in OpenAI provider
from flyto_pro_core.interfaces.providers.openai_llm import OpenAILLMService
llm = OpenAILLMService(model="gpt-4o")
response = await llm.generate("Hello")

# Or implement your own
class MyLLM(ILLMService):
    async def generate(self, prompt, **kwargs) -> LLMResponse:
        ...

Agent Runtime — Verification & Project State

from flyto_pro_core.agent_runtime.verification import DeterministicVerifier
from flyto_pro_core.agent_runtime.project import ProjectStateManager

# Deterministic verification (no LLM needed)
verifier = DeterministicVerifier()
report = await verifier.verify(assertions, evidence)

# Project state management
state = ProjectStateManager(project_dir="/path/to/project")
await state.initialize()

DI Container

from flyto_pro_core.core.container import container

# Register services
container.register("llm", my_llm_instance)
container.register_factory("vector_store", lambda: QdrantVectorStore())

# Retrieve
llm = container.get("llm")

Factory - Compose A Workflow

from flyto_blueprint import BlueprintEngine
from flyto_blueprint.storage.memory import MemoryBackend
from flyto_pro_core.factory import generate_v2

blueprints = BlueprintEngine(storage=MemoryBackend())
result = await generate_v2(
    "Fetch an API and save it to a file",
    blueprint_engine=blueprints,
)
if result.ok:
    print(result.steps)
else:
    print(result.error)

Architecture

flyto-pro-core (this package, Apache-2.0)
├── contract/        → WorkflowSpec → ValidationReport → ExecutablePlan
├── cost/            → BudgetConfig → CostController → BudgetExceededError
├── interfaces/      → ILLMService / IVectorStoreRepository / IQualityChecker
│   └── providers/   → OpenAILLMService, QdrantVectorStore (built-in)
├── agent_runtime/   → Verification, Observations, ProjectState, Interventions
├── core/            → ServiceContainer, safe_access, validators
└── config/          → Settings, constants

Relationship to Flyto2 Ecosystem

flyto-pro-core (open source)     flyto-pro (proprietary)
├── contract                      ├── ems (error learning)
├── cost                          ├── evolution (module generation)
├── interfaces                    ├── knowledge (semantic search)
├── agent_runtime                 ├── agent (AI agent core)
├── core                          ├── guardian (safety)
└── config                        └── enterprise runtime extensions
         │                                    │
         └────────── flyto-ai ────────────────┘
                  (open source)
                  ProBridge connects both layers

• Free users: flyto-ai + flyto-pro-core = full contract validation, cost control, agent runtime
• Pro users: + flyto-pro = EMS error learning, module evolution, semantic knowledge search

Requirements

• Python 3.10+
• pydantic >= 2.0.0
• pyyaml >= 6.0

Optional:

• openai >= 1.0.0 (for OpenAI provider)
• qdrant-client >= 1.7.0 (for Qdrant provider)
• flyto-core >= 2.26.9, < 3 (for contract catalog loading)
• flyto-blueprint >= 0.2.1, < 0.3 (for Factory composition)

API Reference

The generated Python API reference inventories all 923 public classes, functions, and methods with signatures, purposes, and source links. Domain guides explain contracts and operational boundaries: contract engine, agent runtime, cost control, providers, and Factory.

Configuration

Start with Configuration and the generated environment reference. .env.example contains safe examples for every literal environment-variable read; credentials remain blank.

Testing

python -m pytest
python -m ruff check .
python scripts/generate-api-reference.py --check
python scripts/generate-config-reference.py --check
python scripts/check-documentation.py
python -m build

The verification boundary and optional live-service exclusions are mapped in Features and Testing.

Contributing

Pull requests are welcome for contract validation, budget controls, provider interfaces, deterministic verification, docs, and examples. Security reports should go to security@flyto2.com.

License

Apache License 2.0 — see LICENSE.