Roadmap¶

FastAgentic targets staged releases that progressively deliver the deployment layer for agentic applications. This roadmap outlines planned milestones across adapters, operations, governance, integrations, and documentation.

Core Mission: Build agents with anything. Ship them with FastAgentic.

Philosophy: FastAgentic owns the deployment layer. For specialized concerns (observability, guardrails, memory, evaluation), we provide fine-grained hooks to integrate best-of-breed solutions rather than reinventing wheels.

Documentation Philosophy: Every feature ships with decision-focused documentation that explains why you should use it and when it applies. Reference docs tell you how; guides tell you why.

Build vs Integrate¶

FastAgentic focuses on deployment-layer concerns. Specialized tools handle their domains better.

First-Class Integrations¶

These integrations are officially documented, tested, and maintained:

Community integrations welcome via plugin system.

Documentation Strategy¶

Documentation is a first-class deliverable in every release. Each feature ships with:

Documentation Index¶

v0.1 — Foundations¶

Status: Shipped

• App lifecycle with ASGI + MCP dual boot
• Decorators: @tool, @resource, @prompt, @agent_endpoint
• Schema fusion between OpenAPI 3.1 and MCP manifests
• LangChain adapter with streaming support
• Server-Sent Events for long-running operations
• Contract testing via fastagentic test contract

Documentation: - Getting Started guide - Why FastAgentic positioning - Architecture overview - Decorator reference - Basic CLI reference

v0.2 — Workflow Durability & Adapter Ecosystem¶

Status: Shipped

Adapters: - PydanticAI adapter with full streaming support - LangGraph adapter with node-level checkpointing - CrewAI adapter with per-agent observability

Protocols: - MCP 2025-11-25 specification compliance - MCP Tasks for long-running operations - A2A v0.3 protocol implementation - Agent Registry for internal agent discovery - Agent Card auto-generation from endpoints

Durability: - Durable store backends (Redis and PostgreSQL) - Resume, replay, and checkpoint inspection endpoints - Idempotency keys and run deduplication - MCP Task state mapping to durable runs

Templates: - fastagentic-templates repository with index.json registry - fastagentic new --template pydanticai - fastagentic new --template langgraph - fastagentic new --template crewai - fastagentic new --template langchain - fastagentic templates list command with remote discovery - Community template contribution workflow

DevOps: - Docker reference image with production defaults - Helm chart beta with basic configuration - Environment variable configuration reference

CLI: - Enhanced scaffolding for agent workflows - fastagentic mcp validate command - fastagentic a2a validate command

Documentation: - Decision Guides: - Choosing an Adapter — Which framework adapter fits your use case - Choosing Integrations — What tools to add for production - Production Checklist — Are you ready to deploy? - Adapter Guides: - Complete guides for PydanticAI, LangGraph, CrewAI, LangChain adapters - Adapter comparison with feature matrix - Migration guides between adapters - Protocol Documentation: - MCP 2025-11-25 implementation guide - A2A v0.3 with Agent Registry - Protocol interoperability guide - Operations: - Docker deployment guide - Kubernetes with Helm - Environment configuration reference

v0.3 — Hooks, Integrations & Governance¶

Status: Shipped

Hooks Architecture: - Fine-grained hook system with lifecycle points: - on_request / on_response — Request/response transformation - on_tool_call / on_tool_result — Per-tool interception - on_node_enter / on_node_exit — LangGraph node-level hooks - on_llm_start / on_llm_end — LLM call instrumentation - on_checkpoint / on_resume — Durability lifecycle - on_error / on_retry — Error handling hooks - Hook registration via decorators and configuration - Async hook execution (non-blocking where possible) - Hook context with run metadata, user info, cost tracking

Observability Integrations: - LangfuseHook — LLM tracing, prompt analytics - LogfireHook — PydanticAI native observability - DatadogHook — APM integration - Built-in OTEL exporter improvements - Per-agent cost attribution in traces

Guardrail Integrations: - LakeraHook — Prompt injection detection, content moderation - GuardrailsAIHook — Output validation with RAIL specs - NeMoGuardrailsHook — Conversational guardrails - Pre/post execution hook points - Fail-open vs fail-closed configuration

Memory Integrations: - MemoryProvider abstract interface - Mem0Provider — Persistent user memory - ZepProvider — Session memory with summaries - RedisProvider — Simple built-in option - Memory injection into agent context

Rate Limiting & Cost Control: - Simple built-in rate limiter (RPM, TPM, by user/tenant) - PortkeyGateway — LLM gateway with fallbacks, caching - LiteLLMGateway — Multi-provider routing - Cost guardrails with automatic model downgrade - Budget alerts and hard limits

Policy Engine: - Quotas and scope-role mapping - Tenant isolation with separate stores - Audit logging for run decisions and model usage

Protocols: - MCP Extensions framework support - MCP OAuth authorization (SEP-991, SEP-1046) - A2A push notifications (webhooks) - A2A signed Agent Cards - External agent registration in registry

Observability (Built-in): - Expanded metrics (token utilization, queue depth, hook latency) - Prometheus-compatible telemetry endpoints - Grafana dashboard templates

DevOps: - Kubernetes Operator for CRD-based agent management - Prometheus ServiceMonitor integration - Alert rules library

Documentation: - Integration Guides (Why & When): - Langfuse — When to use, configuration, best practices - Lakera — Prompt injection defense strategies - Mem0 — Long-term memory patterns - Portkey — When to use gateway vs built-in reliability - Guardrails AI — Output validation patterns - Reference Documentation: - Hooks Architecture — Complete hook lifecycle reference - Reliability Patterns — When to use built-in vs external tools - Memory Providers — Provider comparison and selection guide - CLI Reference — Complete command reference - Decision Guides: - "When to use Portkey vs built-in reliability" - "Mem0 vs Zep vs Redis: Memory provider selection" - "Observability tool comparison" - Operations: - Grafana dashboard setup - Alert rules documentation - Troubleshooting runbook

v0.4 — Human-in-the-Loop & Evaluation¶

Status: Shipped

HITL Workflows: - interrupt() checkpoint for human approval - Async approval via webhooks (Slack, Email, custom) - HumanLayerHook — Multi-channel approval routing - Three-way decision model (approve / edit / reject) - Approval timeout and escalation policies - WebSocket-based interactive sessions

Evaluation Integrations: - EvalHook abstract interface - BraintrustHook — Experiment tracking, scoring - LangSmithHook — Trace-based evaluation - MaximHook — Production eval pipelines - Custom LLMJudge for inline evaluation - Soft failure thresholds (don't hard-fail on borderline) - Eval results in traces and dashboards

Prompt Management Integrations: - PromptProvider abstract interface - PromptLayerProvider — Versioning, A/B testing - LatitudeProvider — Prompt CMS - AgentaProvider — Prompt + eval workflow - Prompt injection into agent context - Version pinning in configuration

Reliability Patterns: - RetryPolicy configuration (max attempts, backoff) - CircuitBreaker for failing dependencies - Timeout enforcement per endpoint - Fallback chains for graceful degradation

Protocols: - A2A gRPC transport support - A2A task streaming (SSE + gRPC) - MCP Sampling with Tools (SEP-1577) - Cross-agent task delegation via A2A

Artifacts: - Artifact storage for generated assets - S3/GCS/Azure Blob providers - Artifact references in run metadata

DevOps: - Blue/green deployment support with traffic shifting - Canary analysis integration (Flagger compatibility) - Production runbook documentation

Documentation: - HITL Guides: - Human-in-the-loop patterns guide - Approval workflow design patterns - Integration with existing approval systems - Evaluation Guides: - Braintrust — Experiment tracking setup - LangSmith integration for LangChain users - Custom evaluation strategies - "When to evaluate: Development vs Production" - Prompt Management: - Prompt versioning best practices - A/B testing prompts guide - Prompt provider comparison - Advanced Operations: - Blue/green deployment guide - Canary release patterns - Production runbook template

v0.5 — Scale and Orchestration¶

Status: Shipped

Scale: - Distributed checkpointing across clusters - Job scheduler integration for batch workflows - Auto-scaling based on queue depth metrics - Multi-region deployment patterns

Enterprise: - Multi-tenant isolation with per-tenant durable stores - Fleet-aware cost and policy management - SSO/SAML enterprise authentication

Advanced Hooks: - Hook composition and chaining - Conditional hook execution - Hook performance monitoring - Hook marketplace (community plugins)

Documentation: - Scale Guides: - Multi-region deployment patterns - Distributed checkpointing guide - Auto-scaling configuration - Enterprise: - Multi-tenant architecture guide - SSO/SAML integration - Fleet management patterns - Advanced Hooks: - Hook composition patterns - Building custom hooks guide - Hook marketplace submission guide

v1.0 — Production Suite¶

Status: Shipped

Python SDK: - FastAgenticClient and AsyncFastAgenticClient for API interaction - Streaming support with StreamEvent and StreamEventType - Automatic retry with exponential backoff - Comprehensive error handling (AuthenticationError, RateLimitError, etc.) - RunRequest, RunResponse, ToolCall, ToolResult models

Compliance & PII: - PIIDetector with built-in patterns for email, phone, SSN, credit cards, IP addresses - PIIMasker for masking and redaction - PIIConfig with allowlists, blocklists, and custom patterns - PIIDetectionHook and PIIMaskingHook for request/response filtering - Configurable confidence thresholds and type filtering

Dashboard & Metrics: - StatsCollector for run tracking and statistics - RunStats, EndpointStats, SystemStats data models - MetricsRegistry with Counter, Gauge, Histogram metrics - PrometheusExporter for Prometheus-compatible output - DashboardAPI with health, metrics, and stats endpoints - Configurable dashboard with authentication support

Production Readiness: - ReadinessChecker with comprehensive production checks - Security checks: auth, HTTPS, secrets management - Reliability checks: timeouts, retries, rate limiting - Observability checks: logging, metrics, health endpoints - Compliance checks: PII detection, audit logging - Custom check support with ReadinessCheck - ReadinessReport with scoring and recommendations

Documentation: - SDK Guide — Python client usage - Compliance Guide — PII detection and masking - Dashboard Guide — Metrics and monitoring - Ops Guide — Production readiness checks

v1.1 — Adapter & Template Ecosystem¶

Status: Shipped

New Adapters: - SemanticKernelAdapter - Microsoft Semantic Kernel functions and agents - AutoGenAdapter - Microsoft AutoGen multi-agent conversations - LlamaIndexAdapter - LlamaIndex query engines, agents, and chat engines - DSPyAdapter and DSPyProgramAdapter - DSPy modules and compiled programs

Community Adapter SDK: - CommunityAdapter base class for custom adapters - SimpleAdapter for wrapping simple functions - AdapterMetadata for adapter registration - AdapterRegistry for discovery and management - @register_adapter decorator

Template Ecosystem: - Template, TemplateMetadata, TemplateVariable, TemplateFile - TemplateVersion with compatibility checking - LocalRegistry for file-based templates - RemoteRegistry for remote template sources - EnterpriseRegistry with access control and auditing

Template Marketplace: - Marketplace for community template discovery - TemplateRating and TemplateReview for feedback - Browse by category, framework, tags - Search, ratings, and reviews

Template Composition: - TemplateComposer for combining templates - CompositionConfig for merge strategies - File conflict detection and resolution - Python, JSON, YAML merge support

Documentation: - New Adapters Guide — SK, AutoGen, LlamaIndex, DSPy - Template Ecosystem Guide — Templates, marketplace, composition

v1.2 — Interactive Agent CLI¶

Status: Shipped

Agent CLI: - Interactive REPL for agent testing (Claude Code / Gemini CLI-like experience) - fastagentic agent chat - Full interactive chat session - fastagentic agent query - Single queries for scripting and piping - fastagentic agent config - Configuration management - fastagentic agent history - Conversation history management

REPL Features: - Streaming responses with real-time updates - Tool call visualization with formatted output - Conversation history with save/load to JSON - Configurable output formats (markdown, plain, json) - Slash commands for configuration and navigation - Server health checking - File input support

Scripting Support: - Stdin/stdout piping for automation - Output file support - Environment variable configuration - Config file at ~/.fastagentic/config.json

Documentation: - Agent CLI Guide — Commands, configuration, examples

Beyond v1.2¶

Integration Ecosystem: - Community hook marketplace - Certified integration program - Integration testing framework - Partner integration guides

Advanced Features: - Multi-modal streaming (audio, image artifacts) - Marketplace for shared prompts, tools, and adapters - Compliance packs for industry-specific regulations (HIPAA, PCI-DSS) - Agent simulation and load testing

Hook Lifecycle Reference¶

Request Flow:
─────────────────────────────────────────────────────────────────
│ on_request │ → │ on_llm_start │ → │ on_tool_call │
─────────────────────────────────────────────────────────────────
                         │                    │
                         ▼                    ▼
              ┌─────────────────┐   ┌─────────────────┐
              │ Guardrail Hooks │   │   Eval Hooks    │
              │ (Lakera, etc.)  │   │ (Braintrust)    │
              └─────────────────┘   └─────────────────┘
                         │                    │
                         ▼                    ▼
─────────────────────────────────────────────────────────────────
│ on_tool_result │ → │ on_llm_end │ → │ on_response │
─────────────────────────────────────────────────────────────────
                              │
                              ▼
                    ┌─────────────────┐
                    │ on_checkpoint   │ (if durable)
                    └─────────────────┘

Error Flow:
─────────────────────────────────────────────────────────────────
│ on_error │ → │ on_retry │ (if retry policy) → │ resume flow │
─────────────────────────────────────────────────────────────────

LangGraph-Specific:
─────────────────────────────────────────────────────────────────
│ on_node_enter │ → │ node execution │ → │ on_node_exit │
─────────────────────────────────────────────────────────────────

Contributing¶

Timelines are subject to change as community feedback and adoption inform priorities. Contributions, RFCs, and plugin proposals are welcome to accelerate the roadmap.

• GitHub Discussions - Feature requests
• GitHub Issues - Bug reports
• Contributing Guide - How to contribute
• Integration Guide - Build a hook or provider