deepstrike 0.2.4

No project description provided

DeepStrike Python SDK

Runtime framework built on a Rust kernel. The kernel handles loop control, context compression, skill routing, governance, signal prioritization — the SDK handles all I/O.

Install

pip install deepstrike

Requires Python 3.10+. The Rust kernel is distributed as a pre-built wheel (deepstrike._kernel).

---

Quick start

import asyncio
from deepstrike import (
    InMemorySessionLog,
    LocalExecutionPlane,
    OpenAIProvider,
    RuntimeOptions,
    RuntimeRunner,
    collect_text,
    tool,
)

@tool
async def add(x: int, y: int) -> str:
    """Add two numbers."""
    return str(x + y)

plane = LocalExecutionPlane().register(add)
runner = RuntimeRunner(RuntimeOptions(
    provider=OpenAIProvider(api_key="sk-...", model="gpt-5-mini"),
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,
    max_turns=25,
))

asyncio.run(collect_text(runner.run_streaming("What is 17 + 28?")))
# => "45"

Streaming:

async for event in runner.run_streaming("Summarize README.md"):
    if isinstance(event, TextDelta):
        print(event.delta, end="", flush=True)
    elif isinstance(event, ToolCallEvent):
        print(f"\n[→ {event.name}]")
    elif isinstance(event, DoneEvent):
        print(f"\ndone in {event.iterations} turns ({event.status})")

---

Providers

Class	Backend	Notes
OpenAIProvider	OpenAI API	SSE tool-call accumulation
AnthropicProvider	Anthropic API	Native SSE, ThinkingDelta support
QwenProvider	DashScope	enable_thinking via extensions
DeepSeekProvider	DeepSeek API	Reasoner models strip tools automatically
MiniMaxProvider	MiniMax API	M1 reasoning via expose_reasoning
OllamaProvider	Local Ollama	http://localhost:11434 default
KimiProvider	Moonshot API

All providers accept RetryConfig for exponential backoff and share a CircuitBreaker.

---

Context model (four slots)

The kernel renders context as four LLM API slots — only history is compressed.

Slot	Source	Role
system_stable	system partition	Identity, rules — never changes within a run
system_knowledge	knowledge partition	Preloaded memory, skill defs — low frequency
turns[0]	task_state + signals	Goal, plan, progress, compression log, runtime signals
turns[1..N]	history	Conversation transcript

runner = RuntimeRunner(RuntimeOptions(
    initial_memory=["User prefers chartreuse."],  # → Slot 2
    system_prompt="You are a helpful assistant.",  # → Slot 1
    # ...
))

Full reference: docs/context-partition-compression.md

---

Runtime options

runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,            # context window size
    max_turns=25,               # max turns (default 25)
    timeout_ms=60_000,          # timeout in ms (None = no limit)
    extensions={"temperature": 0.1},
    skill_dir="./skills",       # skill .md files directory
    knowledge_source=my_ks,     # KnowledgeSource implementation
    governance=gov,             # kernel Governance instance
    signal_source=gateway,      # SignalGateway or any SignalSource
    dream_store=my_store,       # DreamStore for long-term memory
    agent_id="my-agent",        # required with dream_store for memory meta-tool
    initial_memory=["..."],     # preloaded blocks → Slot 2 (system_knowledge)
    sub_agent_harness=SubAgentHarnessConfig(  # optional: HarnessLoop for spawned sub-agents
        eval_provider=eval_provider,
        max_attempts=3,
    ),
))

---

Tools

from deepstrike import tool, read_file

plane.register(tool(name="search", description="Search.", parameters=schema)(my_fn))
plane.register(read_file)
plane.unregister("search")

---

Skills

Set skill_dir — the kernel auto-injects a skill meta-tool, and the LLM loads skills by name on demand.

runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,
    max_turns=25,
    skill_dir="./skills",
))

---
name: summarize
description: Summarize text into 2-3 concise bullet points
when_to_use: When you need to condense long text
effort: 1
---
1. Identify the 2-3 most important points
2. Express each as a concise bullet

---

Knowledge

Implement KnowledgeSource — the kernel injects a knowledge meta-tool. Runtime retrieval results land in history as tool results. Use initial_memory for durable preload into Slot 2.

from deepstrike import KnowledgeSource

class VectorSearch(KnowledgeSource):
    async def init(self) -> None:
        await vector_db.connect()

    async def retrieve(self, query: str, top_k: int = 5) -> list[str]:
        return await vector_db.search(query, top_k)

runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,
    max_turns=25,
    knowledge_source=VectorSearch(),
))

---

Memory

WorkingMemory (SDK-side scratch pad)

WorkingMemory is an SDK helper — not the kernel working partition (removed). Kernel task state renders into Slot 3 (turns[0]).

from deepstrike import WorkingMemory

mem = WorkingMemory()
mem.set("step", 1)
mem.get("step")  # 1
mem.clear()

DreamStore (long-term memory + dreaming pipeline)

from deepstrike import DreamStore

class MyStore(DreamStore):
    async def load_sessions(self, agent_id): ...
    async def load_memories(self, agent_id): ...
    async def commit(self, agent_id, result, existing): ...
    async def search(self, agent_id, query, top_k): ...
    async def save_session(self, data): ...

runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,
    max_turns=25,
    dream_store=MyStore(),
    agent_id="my-agent",
))

# In-session: LLM calls memory(query) → DreamStore.search() → history tool result
# Preload:    initial_memory → Slot 2 (system_knowledge)
# Post-session: trigger memory consolidation
result = await runner.dream("my-agent", now_ms=int(time.time() * 1000))

---

Governance

SDK PermissionManager

from deepstrike import PermissionManager, PermissionMode

pm = PermissionManager(PermissionMode.DEFAULT)
pm.grant("fs", "read")
pm.revoke("db", "drop")
pm.evaluate("fs", "read")  # PermissionDecision(allowed=True, ...)

Kernel Governance (full pipeline)

from deepstrike import Governance

gov = Governance("allow")
gov.add_permission_rule("danger.*", "deny")
gov.block_tool("rm_rf")
gov.set_rate_limit("api_call", max_calls=10, window_ms=60_000)

runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,
    max_turns=25,
    governance=gov,
))
# Every tool call: Permission → Veto → RateLimit → Constraint → Audit

---

Signals

from deepstrike import SignalGateway, ScheduledPrompt, RuntimeSignal

gw = SignalGateway()

gw.schedule(ScheduledPrompt(goal="standup", run_at_ms=target_time))
gw.ingest(RuntimeSignal(kind="interrupt", payload={}, urgency="critical"))

runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    max_tokens=4096,
    max_turns=25,
    signal_source=gw,
))

runner.interrupt()  # direct interrupt
gw.destroy()

---

Harness (evaluation framework)

from deepstrike import SinglePassHarness, EvalLoopHarness, HarnessLoop, HarnessRequest

# 1. SinglePass — run once, always passes
outcome = await SinglePassHarness(runner).run(HarnessRequest(goal="Say hello"))

# 2. EvalLoop — retry until QualityGate passes
class ContainsHello(QualityGate):
    async def evaluate(self, request, outcome) -> bool:
        return "hello" in outcome.result.lower()

outcome = await EvalLoopHarness(runner, gate=ContainsHello(), max_attempts=3).run(req)

# 3. HarnessLoop — LLM-as-judge with feedback injection + skill extraction
loop = HarnessLoop(runner, eval_provider=eval_provider, max_attempts=3, skill_dir="./skills")

# Sub-agents: pass sub_agent_harness on RuntimeOptions to auto-evaluate spawned children
from deepstrike import SubAgentHarnessConfig
runner = RuntimeRunner(RuntimeOptions(
    provider=provider,
    session_log=InMemorySessionLog(),
    execution_plane=plane,
    sub_agent_harness=SubAgentHarnessConfig(eval_provider=eval_provider, max_attempts=3),
    # ...
))
async for event in loop.run_streaming(HarnessRequest(goal="Write a haiku")):
    if event.type == "done":
        print(event.verdict.passed, event.verdict.feedback)

---

Stream events

Class	Key fields
TextDelta	delta
ThinkingDelta	delta
ToolCallEvent	id, name, arguments
ToolDeltaEvent	call_id, name, delta, chunk?
ToolSuspendEvent	call_id, name, suspension_id, payload?
ToolResultEvent	call_id, content, is_error
DoneEvent	iterations, total_tokens, status
ErrorEvent	message

status: completed · max_turns · token_budget · timeout · user_abort · error