Skip to content

Roadmap

What's been built, what's next, and where AgentTel is heading.

Current Release: v0.3.0-alpha

Everything below is implemented and available.

Python SDK (agenttel-python)

  • Full feature parity with JVM SDK, packaged as agenttel on PyPI
  • Core: semantic attributes, Pydantic models, YAML config, rolling baselines, anomaly detection, SLO tracking, error classification, causality tracking
  • FastAPI integration: middleware, @agent_operation decorator, auto-configuration
  • GenAI instrumentation: OpenAI, Anthropic, LangChain, AWS Bedrock wrappers with cost calculation
  • Agent interface: MCP server (JSON-RPC 2.0) with 15 tools, RBAC, health aggregation, incident context, remediation
  • Agentic observability: AgentTracer with 10 scope context managers, orchestration patterns, cost/quality tracking
  • 74 tests across 9 test files

Previous Release: v0.2.0-alpha

Core Instrumentation

  • Multi-module Gradle project (Java 17+, CI on JDK 17 + 21)
  • agenttel-api — annotations (@AgentOperation, @AgentObservable, @DeclareDependency), attribute constants, enums
  • agenttel-core — span enrichment with topology, baselines (static + rolling + composite), decisions, anomaly detection (z-score), pattern matching, SLO tracking, structured events via OTel Logs API
  • agenttel-spring-boot-starter — auto-configuration, annotation scanning, AOP
  • agenttel-javaagent — zero-code OTel javaagent extension (no Spring dependency)
  • YAML config-driven operations with operation profiles
  • Topology as OTel Resource attributes (set once per service)

GenAI Instrumentation

  • LangChain4j: tracing wrappers for chat, streaming, embedding, and RAG content retrieval
  • Spring AI: span enrichment for existing Micrometer spans
  • Anthropic SDK, OpenAI SDK, AWS Bedrock: tracing wrappers
  • Per-model cost calculation (gen_ai.usage.cost_usd)
  • Auto-configuration with @ConditionalOnClass detection

Agent Interface (MCP Server)

  • 15 MCP tools over JSON-RPC: health, incidents, remediation, SLO reports, trends, executive summaries, cross-stack context, playbooks, verification, error analysis, sessions
  • AgentContextProvider as single entry point for all agent queries
  • Structured incident packages (what's happening, what changed, what's affected, what to do)
  • Remediation execution with action tracking as OTel spans
  • Error classification into actionable categories (dependency_timeout, code_bug, rate_limited, etc.)
  • Causality tracking with change correlation
  • Machine-readable playbooks (cascade failure, error rate spike, latency degradation, memory leak)
  • Action feedback loop with pre/post health verification

Multi-Agent Support

  • Role-based tool permissions (observer, diagnostician, remediator, admin)
  • Agent identity extraction from HTTP headers and tool arguments
  • Shared incident sessions with blackboard pattern
  • Support for coordinator, parallel, swarm, and hierarchical orchestration patterns

Frontend Telemetry

  • @agenttel/web TypeScript browser SDK
  • Auto-instrumentation: page loads, SPA navigation, API calls, clicks, errors
  • Journey tracking with funnel detection and abandonment analysis
  • Client-side anomaly detection: rage clicks, API failure cascades, slow loads, error loops
  • W3C Trace Context cross-stack correlation
  • Per-route baselines and decision metadata

Developer Tooling

  • agenttel-instrument — Python MCP server for IDE-based instrumentation
  • Codebase analysis, config generation, validation, and auto-improvements
  • agenttel-testing — in-memory collector, fluent assertions, JUnit 5 extension
  • 274 tests across all modules

Next: v0.2.0-alpha

Agent Layer

  • [ ] MCP Server authentication (API key, mTLS)
  • [ ] Server-Sent Events (SSE) transport for real-time streaming
  • [ ] WebSocket transport option
  • [ ] Rate limiting on MCP tool invocations
  • [ ] Remediation execution hooks (webhook, Kubernetes API, AWS API)

Observability

  • [ ] Dependency health state machine with transition events
  • [ ] Multi-service correlation — cross-service incident grouping
  • [ ] Custom metric emission alongside span enrichment

GenAI

  • [ ] Prompt/response content capture (opt-in, with PII redaction)
  • [ ] Token budget tracking and alerting
  • [ ] Model performance comparison spans
  • [ ] Guardrail integration for content safety monitoring

Planned: v0.3.0-beta

Framework Ecosystem

  • [ ] Quarkus extension
  • [ ] Micronaut module
  • [ ] Ktor plugin
  • [ ] Kotlin DSL for idiomatic configuration

Infrastructure

  • [ ] OTel Collector processor (Go) for server-side enrichment
  • [ ] Kubernetes operator for automatic MCP server deployment
  • [ ] Helm chart for MCP server sidecar pattern

Quality

  • [ ] Performance benchmarks (enrichment overhead per span)
  • [ ] Load testing with sustained high throughput
  • [ ] Security audit of MCP server

Planned: v1.0.0

Stability

  • [ ] API stability guarantee — no breaking changes without major version bump
  • [ ] Semantic versioning commitment
  • [ ] Long-term support (LTS) release

Standards

  • [ ] OpenTelemetry community proposal for agenttel.* conventions
  • [ ] OTel GenAI SIG collaboration on gen_ai.* attribute stabilization
  • [ ] CNCF project submission

Documentation & Ecosystem

  • [ ] Interactive tutorial / playground
  • [ ] Video walkthroughs
  • [ ] Migration guides from existing instrumentation
  • [ ] Production deployment guide with reference architectures
  • [ ] Grafana plugin for AgentTel-specific visualizations
  • [ ] VS Code extension for annotation assistance

Compatibility Matrix

Backend (JVM)

AgentTel Java OTel SDK Spring Boot LangChain4j Spring AI
0.2.0-alpha 17, 21 1.59.0+ 3.4.x 1.0.0+ 1.0.0+
0.2.0-alpha 17, 21 1.60.0+ 3.4.x, 3.5.x 1.0.0+ 1.0.0+
0.3.0-beta 17, 21, 25 1.62.0+ 3.x 1.x 1.x
1.0.0 17+ 1.x 3.x 1.x 1.x

Frontend (Browser)

AgentTel TypeScript Browsers
0.2.0-alpha 4.7+ Chrome, Firefox, Safari, Edge (ES2020+)

Backend (Python)

AgentTel Python OTel SDK FastAPI
0.3.0-alpha 3.11+ 1.20.0+ 0.100.0+

Tooling

AgentTel Python (instrument agent)
0.2.0-alpha 3.11+