meshflow — Public API Reference

Complete index of everything exported from import meshflow. All symbols listed here are stable under semantic versioning as of v1.0.0.

Agent Creation

Symbol	Description
Agent	Declarative agent builder — the primary entry point
StructuredAgent	Agent with Pydantic output schema enforcement
token_budget	Context manager for token-optimized multi-step runs

Multi-Agent Patterns

Symbol	Description
Team	Multiple agents with a coordination pattern
MessageBus	Async pub/sub between any agents
GroupChat	AutoGen-style round-robin/auto multi-agent chat
GroupChatManager	Orchestrates GroupChat speaker selection
ConversationResult	Output of a GroupChat session
Supervisor	One orchestrator → N worker agents
SupervisorResult	Output of Supervisor.run()
AdversarialTeam	Proposer → Attacker → Judge debate pattern
AdversarialResult	Output of AdversarialTeam.run()
AgentSession	Stateful multi-turn conversation with history compression
SessionResult	Output of AgentSession.chat()
Turn	Single turn in an AgentSession
AgentPool	Async queue with round-robin agent dispatching
PoolStats	AgentPool utilization statistics

Task + Crew (CrewAI-compatible)

Symbol	Description
Task	Work unit with role assignment and {placeholder} substitution
TaskOutput	Output of a completed Task
Crew	Team of agents running Tasks via a Process
CrewOutput	Output of Crew.kickoff()
Process	Execution strategy: sequential / parallel / hierarchical

Typed State Graph (LangGraph-compatible)

Symbol	Description
StateGraph	Typed workflow graph with conditional edges
END	Terminal node sentinel
START	Entry node sentinel
Channel	State channel descriptor
add	Reducer: append to list
last	Reducer: keep last value
first	Reducer: keep first value
node	Decorator to define a graph node function
interrupt	Pause execution for human approval
Command	Resume command after HITL interrupt
Interrupt	Raised when execution is paused
Send	Route to a specific node with payload
MemorySaver	In-memory graph checkpoint backend
SqliteSaver	SQLite graph checkpoint backend

Flows — Event-Driven Decorator API

Symbol	Description
Flow	Event-driven workflow container
FlowState	Pydantic state model for flows
FlowResult	Output of flow execution
flow_start	Decorator: entry point of a flow
flow_listen	Decorator: react to an event
flow_router	Decorator: conditional routing

Declarative Config

Symbol	Description
MeshFlowConfig	Global configuration model
load	Load config from YAML file path
loads	Load config from YAML string

Tool Ecosystem

Symbol	Description
Tool	Callable tool with name, description, risk tier
ToolRegistry	Registry of all available tools
tool	Decorator to register a function as a Tool
global_registry	Module-level default ToolRegistry
GovernedToolRegistry	Tool registry with permissions and audit trail
ToolPermission	Permission rule for a governed tool
ToolAuditEntry	Audit log entry for tool dispatch
PermissionDeniedError	Raised when tool access is denied
ToolNotFoundError	Raised when tool name is not registered

Governance

Symbol	Description
Policy	Governance policy attached to agents/workflows
PolicyMode	Standard / legal-critical / etc.
policy_for_mode	Construct a Policy for a named mode
HumanInLoopConfig	HITL checkpoint configuration
AgentRole	Enum: planner / researcher / executor / critic / orchestrator / guardian
RiskTier	Enum: READ_ONLY / INTERNAL / EXTERNAL_IO / DESTRUCTIVE
RunStatus	Enum: running / completed / failed / paused
RunResult	Result of a governed workflow run
Evidence	Evidence block attached to an Intent
Intent	Declared action before execution
Message	Inter-agent message
ComplianceProfile	Named compliance ruleset (HIPAA/SOX/GDPR/PCI/NERC)
compliance_profile	Look up a ComplianceProfile by name
list_profiles	List all available compliance profile names

Ledger + Audit

Symbol	Description
ReplayLedger	Append-only governed step ledger
RunDiff	Structured diff between two ledger runs
LedgerBackend	Protocol for custom ledger backends
SQLiteLedgerBackend	SQLite ledger backend
PostgresLedgerBackend	PostgreSQL ledger backend
S3LedgerArchiveBackend	S3 archive backend
LedgerArchiveResult	Result of archiving a run to S3
StepRuntime	15-step governed execution kernel
RuntimeOutcome	Outcome of a StepRuntime execution

Evaluation

Symbol	Description
EvalSuite	YAML-driven evaluation harness
EvalScenario	Single evaluation scenario
EvalResult	Aggregate result of a suite run
ScenarioResult	Result of a single scenario
run_eval	Run an EvalSuite against an agent
EvalBaseline	Saved baseline for regression comparison
BaselineDiff	Diff between current run and baseline
LLMJudge	LLM-as-judge with structured scoring
JudgeScore	Score from LLMJudge
JudgeSuiteResult	Aggregate judge scores
ConversationEval	Multi-turn conversation evaluator
ConversationCase	Test case for ConversationEval
EvalTurn	Turn in a ConversationCase
EvalConversationResult	Result of ConversationEval
TurnResult	Result of a single eval turn
ABTest	A/B test between two agents or prompts
ABVariant	One variant in an ABTest
ABTestResult	Aggregate ABTest result
ABTurnResult	Per-turn result in an ABTest
QualityGate	Block deploys on quality regression
QualityReport	QualityGate evaluation report
FeedbackRecord	Human feedback record
FeedbackStore	Store for FeedbackRecord entries
ShadowResult	Result of a shadow run
shadow_run	Run an agent in shadow mode against production
RegressionAlert	Alert from regression detection
RegressionDetector	Detects regressions across shadow runs

Memory

Symbol	Description
AgentMemory	4-tier memory: Working → Episodic → Semantic → Procedural
MemoryItem	Single memory entry
VectorStore	In-process text embedding store
KnowledgeSource	Single knowledge source (file / text / VectorStore)
AgentKnowledge	Multi-source knowledge retriever
MemoryBackend	Protocol for persistent memory backends
InMemoryBackend	In-memory (non-persistent) backend
SQLiteMemoryBackend	SQLite memory backend
PostgresMemoryBackend	PostgreSQL memory backend
snapshot_from_memory	Export memory to serializable snapshot
restore_memory	Restore memory from snapshot
MemoryConsolidator	Compress and deduplicate memory
ConsolidationReport	Report from memory consolidation
TeamWorkspace	Shared memory across a team
WorkspaceSummary	Summary of team workspace state
SemanticMemoryEntry	Dense-embedded memory entry
SemanticSearchResult	Result of semantic memory search
SemanticMemoryStore	Dense embedding memory store
CrossSessionMemoryStore	Persist memories across sessions
CrossSessionEntry	Entry in CrossSessionMemoryStore

LLM Providers

Symbol	Description
LLM	Universal entry point: LLM("model-name")
AnthropicProvider	Anthropic Claude (with prompt caching)
OpenAICompatibleProvider	OpenAI or any OpenAI-compatible endpoint
GeminiProvider	Google Gemini
BedrockProvider	AWS Bedrock
AzureOpenAIProvider	Azure OpenAI
OllamaProvider	Local Ollama (no API key)
LiteLLMProvider	100+ models via LiteLLM
EchoProvider	Offline echo provider (MESHFLOW_MOCK=1)
AzureIdentityProvider	Azure Managed Identity
BedrockIAMProvider	AWS IAM role-based Bedrock access
VertexAIProvider	Google Vertex AI
ProviderRouter	Route by role × budget × compliance → model
auto_provider	Auto-pick provider from environment
auto_model	Auto-pick model from environment
model_to_provider	Infer provider from model name string
provider_for	Factory: provider_for("ollama", model="llama3.2")
auto_detect_provider	Auto-pick based on available keys/services
PROVIDER_NAMES	List of all supported provider name strings
ModelHealthTracker	Rolling-window per-model health tracking
ModelHealthSummary	Health summary for a model
get_health_tracker	Get the global ModelHealthTracker singleton

Streaming

Symbol	Description
StreamChunk	Token chunk emitted during streaming
BackpressureQueue	Async queue with configurable backpressure
BackpressureStrategy	Drop / block / error strategy
StreamMultiplexer	Fan-out a stream to multiple subscribers
Subscription	A subscriber in a StreamMultiplexer
PartialStructuredOutput	Streaming partial Pydantic model output
PartialOutputChunk	Chunk of a partial structured output
stream_structured	Stream and parse structured output incrementally
RunStreamHub	Per-run stream hub for SSE delivery
get_run_hub	Get or create a RunStreamHub
reset_run_hub	Reset the global RunStreamHub

Guardrails + Security

Symbol	Description
Guardrail	Base protocol for all guardrails
GuardrailResult	Result of a guardrail check
GuardrailStack	Ordered stack of guardrails
GuardrailViolation	Exception raised on strict guardrail failure
PIIBlockGuardrail	Block PII in input or output
ConfidenceGuardrail	Require minimum confidence score
LengthGuardrail	Enforce min/max length
ToxicityGuardrail	Block toxic content
JSONSchemaGuardrail	Enforce JSON schema on output
RegexGuardrail	Require or forbid regex pattern
KeywordBlockGuardrail	Block specific keywords or phrases
CostCapGuardrail	Block if cost exceeds per-call budget
CustomGuardrail	User-defined guardrail function
SensitiveDataDetector	23 PHI/PII + credential pattern detector
SensitiveMatch	Match from SensitiveDataDetector
get_sensitive_detector	Get the global SensitiveDataDetector
PromptInjectionDetector	Detect prompt injection attempts
InjectionMatch	Match from PromptInjectionDetector
InjectionResult	Result of injection scan
PromptInjectionGuardrail	Guardrail wrapping PromptInjectionDetector
SecretScanner	Detect exposed secrets and credentials
SecretMatch	Match from SecretScanner
SecretScanResult	Result of secret scan
SecretScanGuardrail	Guardrail wrapping SecretScanner

Observability

Symbol	Description
EventProjector	Project ledger events into 4 views
AuditTrailProjection	Audit trail event projection
NodeLatencyProjection	Per-node latency tracking
NodeLatencyStats	p50/p95/p99 latency statistics
PolicyViolationProjection	Policy violation event projection
WorkflowSummaryProjection	Workflow summary event projection
WorkflowSummary	Summary of a workflow run
OTELExporter	OTLP/HTTP span exporter
get_global_exporter	Get the global OTELExporter
set_global_exporter	Set the global OTELExporter
reset_global_exporter	Reset to no-op exporter
MetricsCollector	Prometheus-compatible metrics
GenAI	OpenTelemetry GenAI semantic convention constants
MF	MeshFlow-specific span attribute constants
GenAISpanRecord	Recorded GenAI span
SpanStore	In-memory span store
configure_telemetry	One-call OTEL setup
get_span_store	Get the global SpanStore
record_agent_step	Record an agent step span
record_handoff	Record an agent handoff span
record_tool_call	Record a tool call span
record_guardrail	Record a guardrail check span
record_healing_attempt	Record a self-healing attempt span
otel_span	Context manager for manual span creation
otel_is_enabled	Check if OTEL telemetry is active
TraceServer	Visual trace studio server
WebhookDelivery	Single webhook delivery record
WebhookRetryQueue	Durable SQLite-backed retry queue
WebhookReliableDeliverer	High-level webhook sender with retry
AlertRule	Metric-threshold alert rule
AlertRecord	Fired alert record
AlertRuleStore	Store for AlertRule definitions
AlertStore	Store for fired AlertRecord entries
AlertEngine	Evaluates metric points against alert rules
MetricPoint	Single metric observation
MetricStore	Time-series metric store

Full list of exports

import meshflow
print(meshflow.__all__)  # complete list of 200+ stable symbols
print(meshflow.__version__)  # "1.0.0"