Agently 4 Documentation¶
Agently 4 is a structured LLM framework focused on controllable outputs, reliable streaming, tool orchestration, and workflow automation via TriggerFlow.
This site is built from the v4 step-by-step examples and refined by the current implementation details.
Why Agently¶
Agently is designed to solve practical problems that appear quickly in real LLM systems:
- Unstable outputs → enforce output formats and parse safely.
- Slow or opaque responses → stream results and expose decision steps.
- Tool chaos → standardized tool schemas, traceable tool calls.
- Multi-step workflows → event-driven orchestration with TriggerFlow.
- Knowledge grounding → integrate knowledge base pipelines without breaking the agent loop.
- Production control → explicit result control, concurrency limits, and runtime data separation.
Architecture in One Page¶
- Agent: prompt builder + model request + response/result cache.
- Response/Result: one request, many views (data/text/meta) without re-requesting.
- Tools: declarative tool registry + tool call tracing.
- TriggerFlow: event-driven graph with runtime data, concurrency, and stream control.
- Auto Loop: planning loop that combines tools, knowledge base, and memo.
What you'll learn¶
- Core agent request/response patterns
- Output format control and streaming
- Tool usage and multi-stage workflows
- TriggerFlow orchestration and advanced execution control
- Auto-loop planning with knowledge base and memory
Tip: All runnable examples live in
examples/step_by_step/.
Agent Cheatsheet (for AI Coding Agents)¶
This is a compact, agent-readable reference extracted from the docs + examples + Auto Loop implementation. It is not intended for humans.
- Download:
Agently Cheatsheet (EN)