Roadmap

What's coming next.

Agentrail is pre-GA. APIs may change. This page is a living document — priorities shift based on community feedback and contributor availability.

Near-Term

GA Readiness

These items must be addressed before a 1.0 stable release. Breaking changes are expected until this phase is complete.

Core Stability

○Finalize public API surface for runtime-core, host, and orchestration

○Lock the AgentrailPlugin contract

○Lock the AgentrailSessionStore interface

○Lock the AgentrailProfile / HostedProfileDefinition contracts

○Stabilize the event type taxonomy (RuntimeEvent, AgentrailHostEvent)

Developer Experience

○Publish @agentrail/testing with official mock utilities (MockLlmClient, MockRuntimeTool)

○Improve create-agentrail-app templates with real LLM setup out of the box

○End-to-end integration test suite for the host layer

Documentation

○Complete all concept, guide, and reference documentation

○Add API reference docs generated from source

✓Launch a documentation site

Mid-Term

Post-GA Improvements

Planned directions after the core is stable. Sequencing will be driven by community usage patterns and contributor availability.

Storage Abstraction

·Define a storage-agnostic SessionStore interface (remove filesystem path leakage)

·Define a storage-agnostic OrchestrationStore interface

·Provide reference implementations for PostgreSQL and SQLite

Scalability

·Support multiple concurrent orchestration runs per manager

·Reduce in-memory state coupling in OrchestrationManager

·Investigate distributed orchestration (multi-process / multi-node)

LLM Provider System

·Replace global singleton LlmProviderRegistry with instance-scoped registries

·Support per-agent and per-tenant provider configuration

·Add provider-level retry and fallback policies

Plugin System

·Add plugin priority ordering

·Add error isolation — one plugin failure does not block others

·Middleware-style request interceptor chain (replace first-match-wins)

·Shared context bag for inter-plugin communication

Sub-Agent Communication

·Structured output types for sub-agent results (beyond outputText: string)

·Typed agent-to-agent message contracts

Observability

·OpenTelemetry trace integration in the agent loop and tool executor

·Structured logging with configurable log levels

·Built-in metrics: request count, token usage, tool execution latency

Not Planned

Out of Scope

To set clear expectations, these are intentionally excluded. Agentrail is a code-first, self-hosted framework.

✕Hosted cloud platform or SaaS offering

✕Built-in billing or usage metering

✕Proprietary model integrations beyond the open provider API

✕GUI-based agent builder — Agentrail is code-first by design

Want to influence priorities?

Contribute to the roadmap.

Open a feature request to discuss an idea, reference this roadmap in your issue for context, or submit a PR for any roadmap item. Community feedback directly shapes what gets built next.

Open a Feature Request →