interloper-agent 0.132.0
Pydantic data-model library importDeclares Markdown package description
SHA-256372874a3c1bf9f050099852ed30b57df0c1f0af1ed5be8b6beafe46484262c41
MaleculeK(Li)H(Db₅)Md₃(Pa₃Pt)
Evidence
1Metadata-Version: 2.3
2Name: interloper-agent
3Version: 0.132.0
4Summary: Interloper AI agent, on pydantic-ai
5Author: Guillaume Onfroy
6Author-email: Guillaume Onfroy <[email protected]>
7Requires-Dist: interloper-core
8Requires-Dist: interloper-db
9Requires-Dist: interloper-toolkit
10Requires-Dist: pydantic-ai-slim[google,anthropic,openai]>=2.51.0
11Requires-Python: >=3.10
12Description-Content-Type: text/markdown
13
14# interloper-agent
15
16The Interloper assistant: one [pydantic-ai](https://ai.pydantic.dev) agent over the
17shared `interloper-toolkit`, served by `interloper-api` and rendered by the app.
1"""Interloper Agent: the assistant, on pydantic-ai, over the shared toolkit."""
2
3from interloper_agent.agent import TURN_LIMITS, build_agent
4from interloper_agent.history import summary_of
5
6__all__ = ["TURN_LIMITS", "build_agent", "summary_of"]
1"""The assistant: one pydantic-ai agent over the toolkit."""
2
3from __future__ import annotations
4
5import datetime
6
7from google.genai.types import ThinkingLevel
8from interloper_toolkit import ToolkitContext
9from pydantic_ai import Agent, DeferredToolRequests, UsageLimits
10from pydantic_ai.capabilities import AbstractCapability, ProcessHistory
11from pydantic_ai.models.anthropic import AnthropicCompaction, AnthropicModelSettings
12from pydantic_ai.models.google import GoogleModelSettings
13from pydantic_ai.models.openai import OpenAICompaction
14from pydantic_ai.settings import ModelSettings
15
16from interloper_agent.history import COMPACTION_THRESHOLD, elide_tool_returns, summarise
17from interloper_agent.prompts import INSTRUCTIONS
18from interloper_agent.toolset import toolset
19
20TURN_LIMITS = UsageLimits(request_limit=40)
21"""What one turn may spend: enough for a long setup flow, a bound on a loop that never converges."""
22
23
24def build_agent(model: str) -> Agent[ToolkitContext, str | DeferredToolRe …
65:14… model has no reliable notion of "now"; without this the relative
66 # timestamps the presentation rules ask for drift to its training data.
67 now = datetime.datetime.now(datetime.timezone.utc)
68 return f"Current date and time: {now:%Y-%m-%d %H:%M} UTC. Compute relative timestamps from it."
69
70 return agent
71
72
73def c …
77:34… ry`), which removes most of a
78 conversation's bulk without a model call. Past a size the older turns
79 are summarised: server-side where the provider compacts natively
80 (Anthropic at :data:`~interloper_agent.history.COMPACTION_THRESHOLD`,
81 OpenAI's Responses API at its own default), by the agent's own
82 …
No evidence locations were recorded for this file. Raw result