File size: 8,224 Bytes
12c2cae
205cdf1
8f2700b
205cdf1
 
8f2700b
205cdf1
 
8f2700b
 
 
205cdf1
 
8f2700b
 
 
 
 
 
 
205cdf1
 
12c2cae
 
205cdf1
12c2cae
205cdf1
12c2cae
8f2700b
 
 
 
 
 
12c2cae
 
 
 
 
8f2700b
 
205cdf1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8f2700b
 
 
 
 
205cdf1
8f2700b
 
205cdf1
8f2700b
 
 
 
 
205cdf1
8f2700b
 
 
 
 
205cdf1
8f2700b
 
205cdf1
8f2700b
 
 
 
 
205cdf1
8f2700b
12c2cae
205cdf1
c110410
 
 
 
 
205cdf1
358e009
 
 
 
 
205cdf1
2404239
 
12c2cae
8f2700b
 
 
 
 
 
 
205cdf1
 
 
 
 
 
 
 
 
 
 
 
 
 
8f2700b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
205cdf1
8f2700b
c110410
 
 
358e009
 
 
2404239
 
12c2cae
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
"""
purpose_agent β€” A local-first self-improvement kernel for agents.

Turns traces into tested memory, policies, and rubrics so agents improve
without fine-tuning, cloud infrastructure, or vendor lock-in.

Architecture based on 13 published papers.
See COMPILED_RESEARCH.md for the full research trace.

Modules:
  Core:       types, llm_backend, actor, purpose_function, experience_replay, optimizer, orchestrator
  V2 Kernel:  v2_types, trace, memory, compiler, immune, memory_ci, evalport, benchmark_v2
  Research:   meta_rewarding, self_taught, prompt_optimizer, llm_compiler, retroformer
  SLM:        slm_backends (Ollama, llama-cpp, prompt compression)
  Streaming:  streaming (async generators, event streaming)
  Tools:      tools (Tool base class, built-in tools, Tool RAG)
  Observe:    observability (cost tracking, callbacks, metrics)
  Multi:      multi_agent (shared memory, agent delegation, teams)
  HITL:       hitl (checkpoint, interrupt, resume, Ξ¦ overrides)
  Eval:       evaluation (benchmark runner, improvement curves)
  Unified:    unified (Agent, Graph, parallel, Conversation, KnowledgeStore)
  Easy:       easy (purpose(), Team, quickstart wizard)
"""

__version__ = "2.0.0"

# ── Core ──────────────────────────────────────────────────────────────────
from purpose_agent.types import (
    State, Action, Trajectory, TrajectoryStep,
    Heuristic, PurposeScore, MemoryRecord, MemoryTier,
)
from purpose_agent.llm_backend import (
    LLMBackend, MockLLMBackend, HFInferenceBackend,
    OpenAICompatibleBackend, ChatMessage,
)
from purpose_agent.actor import Actor
from purpose_agent.purpose_function import PurposeFunction
from purpose_agent.experience_replay import ExperienceReplay
from purpose_agent.optimizer import HeuristicOptimizer
from purpose_agent.orchestrator import Orchestrator, Environment, SimpleEnvironment, TaskResult

# ── V2 Kernel ─────────────────────────────────────────────────────────────
from purpose_agent.v2_types import RunMode, MemoryScope, PurposeScoreV2
from purpose_agent.trace import Trace, TraceEvent
from purpose_agent.memory import MemoryStore, MemoryCard, MemoryKind, MemoryStatus
from purpose_agent.compiler import PromptCompiler, CompiledPrompt
from purpose_agent.immune import scan_memory, ScanResult
from purpose_agent.memory_ci import MemoryCI
from purpose_agent.evalport import EvalCase, EvalPort, DictEvalPort, ScoreBundle
from purpose_agent.benchmark_v2 import BenchmarkRunnerV2, V2BenchmarkResult

# ── Research Implementations ──────────────────────────────────────────────
from purpose_agent.meta_rewarding import MetaRewardingLoop
from purpose_agent.self_taught import SelfTaughtEvaluator
from purpose_agent.prompt_optimizer import PromptOptimizer, Signature, Demonstration
from purpose_agent.llm_compiler import LLMCompiler, ExecutionPlan, TaskNode
from purpose_agent.retroformer import Retroformer

# ── SLM-Native Backends ──────────────────────────────────────────────────
from purpose_agent.slm_backends import (
    OllamaBackend, LlamaCppBackend, SLMPromptCompressor,
    create_slm_backend, SLM_REGISTRY,
)

# ── Streaming & Async ────────────────────────────────────────────────────
from purpose_agent.streaming import StreamingMixin, StreamEvent, AsyncOrchestrator

# ── Tools ────────────────────────────────────────────────────────────────
from purpose_agent.tools import (
    Tool, FunctionTool, ToolResult, ToolRegistry,
    CalculatorTool, PythonExecTool, ReadFileTool, WriteFileTool,
)

# ── Observability ────────────────────────────────────────────────────────
from purpose_agent.observability import (
    CostTracker, TokenUsage, CallbackManager,
    AgentEvent, EventType, LoggingCallback, MetricsCollector,
)

# ── Multi-Agent ──────────────────────────────────────────────────────────
from purpose_agent.multi_agent import AgentSpec, AgentTeam

# ── Human-in-the-Loop ───────────────────────────────────────────────────
from purpose_agent.hitl import (
    HITLOrchestrator, Checkpoint, HumanInputHandler,
    CLIInputHandler, AutoApproveHandler, InterruptType,
)

# ── Evaluation (V1 compat) ──────────────────────────────────────────────
from purpose_agent.evaluation import BenchmarkTask, BenchmarkRunner, BenchmarkResult

# ── Plugin Registry ──────────────────────────────────────────────────────
from purpose_agent.registry import (
    PluginRegistry, backend_registry, callback_registry, model_registry,
    EmbeddingBackend, default_embedding,
)

# ── Unified Capabilities ────────────────────────────────────────────────
from purpose_agent.unified import (
    Agent, Graph, parallel, Conversation, KnowledgeStore,
    START, END, Message,
)

# ── Easy API (the only thing beginners need) ─────────────────────────────
from purpose_agent.easy import purpose, Team, quickstart, TEAM_TEMPLATES

__all__ = [
    # Core
    "State", "Action", "Trajectory", "TrajectoryStep", "Heuristic",
    "PurposeScore", "MemoryRecord", "MemoryTier",
    "LLMBackend", "MockLLMBackend", "HFInferenceBackend",
    "OpenAICompatibleBackend", "ChatMessage",
    "Actor", "PurposeFunction", "ExperienceReplay", "HeuristicOptimizer",
    "Orchestrator", "Environment", "SimpleEnvironment", "TaskResult",
    # V2 Kernel
    "RunMode", "MemoryScope", "PurposeScoreV2",
    "Trace", "TraceEvent",
    "MemoryStore", "MemoryCard", "MemoryKind", "MemoryStatus",
    "PromptCompiler", "CompiledPrompt",
    "scan_memory", "ScanResult",
    "MemoryCI",
    "EvalCase", "EvalPort", "DictEvalPort", "ScoreBundle",
    "BenchmarkRunnerV2", "V2BenchmarkResult",
    # Research
    "MetaRewardingLoop", "SelfTaughtEvaluator",
    "PromptOptimizer", "Signature", "Demonstration",
    "LLMCompiler", "ExecutionPlan", "TaskNode",
    "Retroformer",
    # SLM
    "OllamaBackend", "LlamaCppBackend", "SLMPromptCompressor",
    "create_slm_backend", "SLM_REGISTRY",
    # Streaming
    "StreamingMixin", "StreamEvent", "AsyncOrchestrator",
    # Tools
    "Tool", "FunctionTool", "ToolResult", "ToolRegistry",
    "CalculatorTool", "PythonExecTool", "ReadFileTool", "WriteFileTool",
    # Observability
    "CostTracker", "TokenUsage", "CallbackManager",
    "AgentEvent", "EventType", "LoggingCallback", "MetricsCollector",
    # Multi-Agent
    "AgentSpec", "AgentTeam",
    # HITL
    "HITLOrchestrator", "Checkpoint", "HumanInputHandler",
    "CLIInputHandler", "AutoApproveHandler", "InterruptType",
    # Evaluation (V1)
    "BenchmarkTask", "BenchmarkRunner", "BenchmarkResult",
    # Plugin Registry
    "PluginRegistry", "backend_registry", "callback_registry", "model_registry",
    "EmbeddingBackend", "default_embedding",
    # Unified Capabilities
    "Agent", "Graph", "parallel", "Conversation", "KnowledgeStore",
    "START", "END", "Message",
    # Easy API
    "purpose", "Team", "quickstart", "TEAM_TEMPLATES",
]