List view
Replace Claude Code CLI agents with custom agents built on the Claude Agent SDK (TypeScript). Gives agentd full control over agent lifecycle, tool execution, and conversation management.
No due date•3/15 issues closedRedesign the ask service from a system-check tool into a purpose-built agent-to-human Q&A system. Agents ask structured questions, humans answer at their convenience, and workflows trigger on responses via the new ask_response trigger type.
No due date•6/16 issues closedAdd experiment tracking for autonomous AI agent runs. Experiments are standalone entities in the orchestrator that directly associate with agents and workflows via join tables — no project dependency. Includes experiment lifecycle management (planned/running/completed/failed/cancelled), dispatch and cost attribution via experiment_id, Prometheus metric labeling, memory auto-tagging, and a local Prometheus/Grafana observability stack. Project scoping deferred to core service (v0.3.3).
No due date•5/12 issues closedIntegrate the Agent Connect Protocol (ACP) into agentd, enabling cross-framework agent interoperability via standardized REST+SSE endpoints for agent discovery, run execution, thread management, and external agent delegation.
No due date•0/17 issues closedFully autonomous AI-driven development pipeline. Includes git spice integration for stacked branch management, conductor agent for process orchestration, specialist agents (refactor, research, tester, triage, enricher, security), merge automation, label-driven state machine, workflow chaining, and updated agent/workflow templates. Proof of concept for agentd as a self-sustaining development platform.
No due date•64/65 issues closedAdd a new communicate service enabling real-time messaging between agents and humans via rooms/channels. Includes service implementation, orchestrator integration, CLI commands, template support, frontend UI, and documentation.
No due date•24/25 issues closedAdd OpenCode as a second agent backend alongside Claude Code. Includes abstracting the agent communication protocol, implementing an HTTP/SSE client for OpenCode's server API, and updating the orchestrator, CLI, and configuration to support multiple agent types.
No due date•0/23 issues closed- No due date•27/44 issues closed
Expand workflow trigger types beyond poll-based GitHub sources. Includes TriggerStrategy abstraction, cron/delay triggers, internal event bus, webhook triggers, and manual dispatch.
No due date•58/59 issues closed- No due date•36/36 issues closed
Port the berry-rs memory system into agentd as the agentd-memory service crate. Adds vector-backed semantic memory with LanceDB, embedding services, REST API, and CLI integration.
No due date•17/17 issues closedReplace tmux-based agent execution with Docker container backend for proper sandboxing, isolation, and resource control. Implements ExecutionBackend trait abstraction, DockerBackend using bollard, Dockerfile for Claude Code, networking, CLI support, and documentation.
No due date•11/11 issues closedImplement the agentd-mcp server providing Model Context Protocol tools for troubleshooting, diagnostics, and self-healing of agentd services.
No due date•12/12 issues closedMigrate all agentd storage layers from raw SQLx to SeaORM, gaining a proper migration framework, entity definitions, and relation support. Eliminates manual row deserialization and inline schema management.
No due date•10/10 issues closedImplement the core service crate (crates/core) providing authentication, user/organization management, tenant isolation, and API gateway functionality. Enables multi-user and multi-tenant support for the agentd system.
No due date•1/23 issues closedFrontend web UI for the agentd project built with React, Vite, Tailwind CSS, TypeScript, Bun, and Nivo. Provides dashboard, agent management, notifications, workflows, monitoring, and real-time streaming.
No due date•20/20 issues closedExtract and persist usage data from every Claude Code result message, expose usage stats via API and CLI, add clear-context action that restarts agent process while preserving UUID, and support optional auto-clear when token threshold is exceeded. Builds on the insight that restart_agent already preserves the agent UUID, making it safe to use as a context-clearing mechanism.
No due date•20/20 issues closedExpand the platform with new task sources, server-side agent name resolution, and comprehensive cross-service integration tests.
No due date•6/12 issues closedOperational concerns for running agentd in real environments: authentication, configuration management, resilient communication, database migrations, sandboxing, and containerized deployment.
No due date•8/26 issues closedImplement the three stub crates (hook, monitor, ollama), integrate BAML, and extract the shared HTTP client. Brings every crate in the workspace to a functional state.
No due date•10/15 issues closedFix known bugs, consolidate duplicated code into agentd-common crate, and document what exists. No new features — making the current codebase solid before building on it.
No due date•21/23 issues closed