Releases: voocel/agentcore
Releases · voocel/agentcore
Release list
v1.5.2
Features
- Implement tool search with defer loading support.
- Add policy configuration.
Performance
- Add reactivate deferred mechanism.
- Add system blocks for enhanced processing.
- Stream reasoning delta for subagents.
- Update policy handling.
Other
- Update CI configuration and add GitHub release action.
- Update README documentation.
v1.5.1
This release focuses on stronger multi-agent execution and runtime stability
It adds multi-modelsupport
background subagents
image-capable read tools
LLM transform/streaming support
while improving parallel tool calls, abort handling, subagent progress reporting
v1.5.0
agentcore v1.5.0
A ground-up rewrite. old version was a heavy orchestration framework with complex workflows
and rigid abstractions. new version strips it all away — what remains is a minimal, composable
library for building AI agents in Go.
What Changed
old version tried to do everything: workflow orchestration, state machines, chained pipelines.
new version does one thing well: a clean agent loop with streaming events.
- Single event stream drives everything — TUI, Web, CLI, whatever you build
- Tool interface is 4 methods, not a framework
- Context pipeline (transform + convert) replaces rigid middleware chains
- Multi-agent via tool invocation, not a separate orchestration layer
Core
Agent— stateful wrapper with subscribe/prompt/abort lifecycleAgentLoop— double-loop architecture (inner: tool calls + steering, outer: follow-ups)Event— 11 event types covering the full agent lifecycleToolinterface — Name, Description, Schema, Execute. That's it.SubAgent— parallel and chained multi-agent execution as a toolStreamFn— bring your own LLM, or use the built-in adapters