lllm/__init__.py |
Public exports, _auto_init() to populate the default runtime from lllm.toml. |
lllm/core/resource.py |
ResourceNode (lazy/eager wrapper), PackageInfo, and convenience loaders (load_prompt, load_tactic, load_proxy, load_config, load_resource). |
lllm/core/runtime.py |
Runtime — unified ResourceNode-based registry with namespace-aware resolution. Named runtimes via load_runtime / get_runtime. |
lllm/core/config.py |
Package loading (load_package), TOML parsing, resource discovery, config inheritance (resolve_config, _deep_merge), AgentSpec, parse_agent_configs. |
lllm/core/prompt.py |
Prompt model — template, parser, tools, handler. Function, FunctionCall, MCP descriptors. |
lllm/core/dialog.py |
Dialog — append-only message sequence with tree structure. Message, DialogTreeNode. Serialization uses qualified prompt keys. |
lllm/core/agent.py |
Agent — dialog management, agent call loop, retry/interrupt handling. |
lllm/core/tactic.py |
Tactic — top-level abstraction wiring agents to prompts. TacticCallSession, _TrackedAgent, build_tactic, register_tactic_class. |
lllm/core/const.py |
Enumerations (Roles, Modalities, APITypes), model cards, pricing utilities. |
lllm/core/log.py |
Replayable logging base classes (ReplayableLogBase, LocalFileLog, NoLog). |
lllm/proxies/base.py |
BaseProxy, ProxyManager runtime, ProxyRegistrator decorator, endpoint metadata. Namespace-aware proxy activation. |
lllm/proxies/builtin/ |
Built-in proxy implementations (financial data, search, etc.). |
lllm/invokers/ |
Provider registry and concrete backends implementing BaseInvoker. |
lllm/sandbox/ |
JupyterSession, kernel management for code execution. |
lllm/tools/cua.py |
Experimental Computer Use Agent (Playwright browser automation). |
lllm/cli.py |
lllm command-line interface and template renderer. |
lllm/utils.py |
Filesystem helpers, caching, HTTP utilities, stream wrappers. |