Picks

What we would read and try first, and why.

Papers

  • Proactive Agent

    The baseline to start with: anticipating tasks from desktop activity, with a benchmark to match.

  • PASK

    Shows the whole decision: stay silent, step in briefly or assist in depth, with interruption cost made explicit.

  • HALO

    Asks what most evaluations skip: did people find the help acceptable, well-timed and the right intensity?

Products

  • OpenAI dot ↗

    A mainstream example of ongoing responsibility: research and checks that continue between conversations.

  • Meta Muse ↗

    A standing goal drives background work, and it returns when conditions change or approval is needed.

  • Grok Bot ↗

    Scheduled and event-triggered routines in product form, including when to escalate to the user.

Open projects

  • OpenClaw ↗

    Read its heartbeat design to see how a runtime checks for meaningful change and suppresses noise.

  • Proactivity SDK ↗

    A compact reference for wake-up logic: durable goals, model-selected cadence and action caps.

  • MineContext ↗

    An open desktop app that turns screen context into tips and todos: an ambient loop you can run yourself.

More papers worth your time

Interaction Models

Agents & Systems

Questions, Preferences & Boundaries

Evaluation

Reading path

A layered starting path, selected for coverage of different mechanisms and evidence, rather than publication date or a shared leaderboard. Read Tier 1 first, choose an application or mechanism in Tier 2, then use Tier 3 to evaluate it. Tiers are a reading order, not a quality ranking. Products and code are included alongside papers.

RouteRepresentative resourcesWhy start here
Tier 1 · Foundations — define proactivity, build a baseline, understand interruption
Human-centered definitionTowards Human-centered Proactive Conversational Agents
SIGIR 2024 · conceptual foundation
Intelligence, adaptivity, and civility provide a starting vocabulary for useful, respectful initiative.
Canonical agent baselineProactive Agent · Code · Notes
ICLR 2025 · method + benchmark
Connect desktop event streams to task anticipation and ProactiveBench evaluation.
User experience and timingNeed Help? · Notes
CHI 2025 · human factors
Understand when proactive IDE help is useful and how users retain control.
Tier 2 · Mechanisms and systems — choose the kind of initiative you want to build
Sensory context → assistanceContextAgent · Code
NeurIPS 2025 · context-aware agent
Move from desktop logs to open-world sensory context and tool-mediated help.
Persistent agent activationOpenClaw · Heartbeat docs
Proactivity SDK
Runtime projects · see Projects & Products
Compare periodic attention checks with durable goals and model-selected wake cadence.
Product-level initiativeOpenAI dot
MineContext
Always-on product / ambient desktop project
Compare background research and connected-app context with proactive desktop summaries and tips.
Latent needs and persistent memoryPASK / IntentFlow · Code / project
Notes
Technical report · need detection and assistance routing
Connect silent / fast / full assistance decisions to ongoing conversations and hierarchical personal memory.
Personal routines across devicesPersonalAlign / HIM-Agent
Personal Agents guide
Mobile user history · preference and routine memory
Separate completing omitted preferences from anticipating an unrequested routine; personal assistance extends beyond desktop interfaces.
Wearable and physical-world helpSatori
ProMemAssist
AR guidance · working-memory-aware intervention
Ground assistance in inferred user/task state and weigh the benefit of help against interruption.
Embodied ask / act decisionsPACT
PACE
Continual robot assistance · action-completion timing
Compare asking versus acting from interaction history with timing physical help from observed task progress.
Visual events → speak / silenceJoyAI-VL-Interaction · Code
MOSS-VL-Realtime · Code
Streaming interaction models
Study internally controlled response timing and vision-triggered intervention, including JoyAI's delegation decision.
Evidence and memory → timely answerOneStreamer · Code · Notes
StreamReady
Streaming memory and answer readiness
Contrast proactive evidence recording with waiting until a standing question has enough evidence.
Full-duplex interaction → agent workMiniCPM-o 4.5 · Code
Gander · Code
Simultaneous perception, speech, and asynchronous work
Study speech/listening control, scene-driven comments, and the bridge to a background reasoning agent.
Tier 3 · Evaluation and boundaries — test the route you chose
Real workflows and personalizationProAgentBench
KnowU-Bench · Code
Pair real desktop traces with personalized, consent-aware mobile evaluation.
GUI intent recommendationPIRA-Bench / PIRF · Project
Data
Benchmark with a memory-aware recommendation baseline
Evaluate latent future intents, user profiles, interleaved tasks, and rejection of noisy visual context before instruction-following execution.
Long-horizon initiativeπ-Bench · Code
VibeLifeBench
Evaluate hidden intents and act / ask / stay-silent decisions over evolving personal workflows.
Streaming timing and intentOmniMMI
StreamGaze · Code
Streaming evaluation routes
Compare multimodal alert/turn timing with gaze-conditioned intent; inspect proactive subsets separately from general streaming QA.
Personalized physical-world assistanceEgoPro-Bench
Vinci2 / EgoServe
Egocentric timing, user context, and silence
Connect streaming to personalized assistance while keeping recorded-video evaluation separate from live wearable deployment.
Silence and oversightWhy2Speak · Notes
Benchmark Matrix
Audit act-versus-abstain decisions, class imbalance, and whether exposed reasoning changes the policy being inspected.