Repository navigation
Roadmap
This page retains historical plans. For shipped behavior use the v1.0.0 guides; for ongoing priorities use the repository's current roadmap. Do not treat the old backlog below as a list of features missing from v1.0.0.
This roadmap keeps LightAgent focused on its core identity:
Lightweight core, composable Skills, reliable tool execution, observable traces, and OpenAI-compatible model support.
Recent work has improved:
- structured run results
- structured streaming events
- catchable and formatted error codes
- tool argument validation
- memory namespace policy
- trace event export
- deterministic LightFlow workflow orchestration
- CI coverage across Python 3.10, 3.11, and 3.12
- contributor, security, and issue templates
Goal: make tool use, memory use, and final responses safer under explicit policy.
Planned work:
- input guardrails for sensitive or disallowed content
- tool-call guardrails for high-risk actions
- approval requirements for file, network, database, payment, and external-action tools
- output guardrails for JSON schema validation and PII checks
- memory guardrails for namespace isolation, provenance checks, and retrieval filtering
Expected outcome: production users can constrain actions and memory behavior without replacing LightAgent's lightweight runtime.
Status: released.
Goal: add deterministic workflow orchestration without turning LightAgent into a heavy framework.
Proposed shape:
flow = LightFlow()
flow.step("research", agent=research_agent)
flow.step("analyze", agent=analysis_agent, depends_on=["research"])
flow.step("write", agent=writer_agent, depends_on=["analyze"])
result = flow.run("Analyze this company")Delivered:
- DAG-style step orchestration
- explicit step input and output passing
- step-level retry policies
- result formats:
object,str, anddict - flow traces using the same trace event system
Deferred to later versions:
- manual approval nodes
- durable execution and resume support
Goal: support high-risk tasks that need explicit review before continuing.
Planned work:
HumanApprovalTool- approval, rejection, and edited arguments
- timeouts and rejection handling
- resume after approval
Goal: make long-running tasks recoverable.
Planned work:
-
RunStoreinterface - SQLite, Redis, and Postgres adapters
- persisted model requests, tool calls, tool results, memory reads, and memory writes
agent.resume(run_id)- idempotency markers for tools
Goal: provide a stable, documented, tested foundation for lightweight production agents.
Planned work:
- stabilize public APIs:
LightAgent,LightSwarm,LightFlow,Skill,ToolRegistry,MemoryProtocol,RunResult - complete documentation site
- example matrix for common providers and workflows
- PyPI release automation
- benchmarks for tool-call success, multi-turn completion, cost, latency, and recovery
Open an issue with:
- the use case
- expected API shape
- safety impact
- testing strategy
- compatibility impact
Use existing priority labels when available: priority:p0, priority:p1, and priority:p2.
LightAgent Wiki - see the repository, releases, and issues.
- Installation and Your First Agent
- Tools and Integrations
- Memory, Sessions, and RAG
- Workflows and Multiple Agents
- Agent Backends and Codex
- Durable Execution, Jobs, and Subagents
- Safety, Observability, and Evaluation
- Usage, Cost Estimates, and Budgets
- Deployment, Migration, and Verification
- 安装与第一个 Agent
- 工具和集成
- 记忆、会话与 RAG
- 工作流与多 Agent
- 执行后端与 Codex
- 持久执行、Jobs 与 Subagents
- 安全、观测与评测
- 用量统计、费用估算与预算
- 部署、升级与验收
- Quick Start
- Core Concepts
- API Reference
- Examples Cookbook
- Migration Guide
- Tools
- Tool Generator
- Memory
- MCP
- Skills
- LightFlow
- Tree of Thought
- Self-Learning
- Multi-Agent
- Tracing and Debugging
- Langfuse Observability
- Model Providers
- browser-use Integration
- Testing and CI
- Deployment Guide
- Architecture
- Security
- Known Limitations
- FAQ
- FAQ 中文
- Roadmap
- Release Process
- Contributing