Skip to content

RUNTIME.md

Latest commit

 

History

History
58 lines (47 loc) · 2.28 KB

RUNTIME.md

File metadata and controls

58 lines (47 loc) · 2.28 KB
Raw

InfiAgent

InfiAgent is a long-horizon multi-agent framework built around three ideas: configuration-defined agents (an agent is YAML, not a class), a forgetting loop (short action windows + periodic thinking summaries + file relay), and a resumable execution stack (all conversation/stack/action state is persisted under a portable user_root/, so tasks survive interruption and resume across processes and machines).

pip install infiagent

This package ships the complete backend runtime: the agent execution loop, context builder, tool executor, the built-in level-0 tool suite, the LLM client with multi-provider deployment failover, the concurrency-safe experience store, and the SDK entry point, together with the full test suite.

Quick start

from infiagent import infiagent

agent = infiagent(
    user_data_root="./user_root",
    agent_library_dir="./user_root",
    default_agent_system="my_system",
    default_agent_name="alpha_agent",
    direct_tools=True,
)
result = agent.run("your task", task_id="./user_root/tasks/demo")

Agent systems are defined under user_root/agent_library/<system>/ (level_3_agents.yaml, level_0_tools.yaml, general_prompts.yaml); model access is configured in user_root/config/llm_config.yaml (single key, or a multi-provider deployments list with sticky-primary failover). Task state persists under user_root/tasks/<task_id>/ and user_root/conversations/, and interrupted tasks can be resumed.

Reliability characteristics

  • Crash-safe persistence: runtime state is written atomically (same-directory temp file + fsync + rename); corrupted legacy state files are quarantined automatically with a clean recovery path, and interrupted tool calls replay idempotently on resume.
  • Concurrency-safe experience store: cross-process file locks over the full read-modify-write, uuid entry ids, file revisions with optimistic concurrency checks, transactional dual-scope writes, and only status=active entries are injected into agent context.
  • Multi-provider LLM failover: a model may declare multiple deployments (keys/platforms); healthy traffic sticks to the primary to preserve provider-side prompt caching, failures cool down and fail over automatically.

License

GPL-3.0. See LICENSE.