ContextPilot-14B

ContextPilot-14B

ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL

ContextPilot-14B is the Qwen3-14B checkpoint of ContextPilot, a proactive context-management framework for long-horizon language-model agents. It teaches agents to plan, maintain long-term memory, and offload less useful context while they continue reasoning and using tools. For more details, see our paper and code repository.

Overview

ContextPilot combines three main components:

• an extended context-management toolset with planning, structured memory, retrieval, and soft context offloading;
• context-aware partial rollout that focuses exploration on sensitive context-editing decisions; and
• fine-grained credit assignment that trains intermediate snapshots using the outcomes of their downstream branches.

The resulting agents are evaluated on long-context question answering and deep-search tasks; see the evaluation instructions for details.

Loading

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "panzs19/ContextPilot-14B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

Note that loading the checkpoint alone does not execute context-management tools; the tool definitions, agent runtime, and evaluation pipeline are provided in the ContextPilot repository. See the inference guide for the full setup.

Intended Use

This checkpoint is intended for research on proactive context management, long-horizon agents, long-context QA, and deep search.