mirror of https://github.com/open-jarvis/OpenJarvis.git synced 2026-07-30 19:02:16 +00:00

Files

T

Jon Saad-FalconandClaude Opus 4.6 8d538cd1b0 Add orchestrator training, channels, LiteLLM engine, and simplify learning taxonomy

Major changes across parallel sessions:

- Add orchestrator SFT & GRPO training subpackage (learning/orchestrator/)
  with episode types, multi-objective reward, prompt registry, policy model,
  RL environment, and registered learning policies
- Add structured THOUGHT/TOOL/INPUT/FINAL_ANSWER mode to OrchestratorAgent
- Add 15 channel backends (Discord, Slack, Telegram, Email, Webhook, IRC,
  Matrix, Teams, WhatsApp, Signal, Mattermost, BlueBubbles, Feishu,
  Google Chat, Webchat) with channel tools and config
- Add LiteLLM engine backend for unified LLM provider access
- Add RLM agent and REPL tool
- Remove ToolLearningPolicy — learning taxonomy now only targets
  Intelligence (LM weights/routing) and Agents (logic/ICL/tool strategies)
- Rename SFTPolicy to SFTRouterPolicy (backward-compat alias kept)
- Remove OpenClaw agent infrastructure (openclaw*.py, openclaw_bridge.py)
- Fix async streaming tests (asyncio.run vs deprecated get_event_loop)
- Fix server channel route tests (pytest.importorskip for optional fastapi)
- Track uv.lock for reproducibility
- Update CLAUDE.md and docs to reflect all changes

1676 tests pass, 37 skipped.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-02-23 18:32:32 +00:00

2.2 KiB

Raw Blame History

Learning Module

The learning module implements learning policies that improve routing, agent, and tool decisions based on historical interaction outcomes. The module provides a LearningPolicy ABC taxonomy with specialized sub-ABCs for intelligence (model routing), agent behavior, and tool selection. It also includes reward functions for scoring inference results.

Abstract Base Classes

RouterPolicy

::: openjarvis.intelligence._stubs.RouterPolicy options: show_source: true members_order: source

QueryAnalyzer

::: openjarvis.intelligence._stubs.QueryAnalyzer options: show_source: true members_order: source

RoutingContext

RoutingContext is defined in core/types.py.

RewardFunction

::: openjarvis.learning._stubs.RewardFunction options: show_source: true members_order: source

LearningPolicy Taxonomy

The learning system defines a hierarchy of learning policy ABCs:

LearningPolicy -- base ABC for all learning policies
IntelligenceLearningPolicy -- specialization for model routing decisions
AgentLearningPolicy -- specialization for agent behavior advice (ICL examples, tool-use strategies)

Policy Implementations

TraceDrivenPolicy

::: openjarvis.learning.trace_policy.TraceDrivenPolicy options: show_source: true members_order: source

classify_query

::: openjarvis.learning.trace_policy.classify_query options: show_source: true

SFTRouterPolicy

::: openjarvis.learning.sft_policy.SFTRouterPolicy options: show_source: true members_order: source

AgentAdvisorPolicy

::: openjarvis.learning.agent_advisor.AgentAdvisorPolicy options: show_source: true members_order: source

ICLUpdaterPolicy

::: openjarvis.learning.icl_updater.ICLUpdaterPolicy options: show_source: true members_order: source

GRPORouterPolicy

::: openjarvis.learning.grpo_policy.GRPORouterPolicy options: show_source: true members_order: source

Reward Functions

HeuristicRewardFunction

::: openjarvis.learning.heuristic_reward.HeuristicRewardFunction options: show_source: true members_order: source

2.2 KiB Raw Blame History