BlogDocs
Sign in

Blog

Agent event infrastructure, without the noise.

Short, practical writing on event layers, delivery channels, OpenClaw and Hermes setup, and making agents proactive without wasteful polling.

Hermes avatar identity tile connected to personal workflow cards, reminders, receipts, and a filtered work queue.
Agent use cases

Real Hermes use cases are weird, personal, and hard to explain

The most durable personal-agent workflows are not generic demos. They are specific, local, recurring, and tied to a user's own tools.

8 min read · Updated 2026-06-02
Personal agent workstation with broken cron tiles, backup drives, and a filtered event queue for safer background work.
Agent reliability

Cron jobs are becoming the first production incident for personal agents

Personal agent users are discovering reliability through cron failures, broken updates, backups, token spikes, and missed background runs.

8 min read · Updated 2026-05-31
Agent demo symbol on a front-stage pedestal with hidden state ledgers, retries, provenance cards, and queues underneath.
Agent production

Agent demos hide the hard part: state

Production agent failures usually appear around state, handoff, memory, retries, vague inputs, and event boundaries rather than one-off model capability.

8 min read · Updated 2026-05-29
Broad operations control desk routing code, incidents, support, billing, docs, chat, inbox, and calendar work into one queue.
Agent use cases

Boring agent workflows are the real ones

The most durable proactive-agent use cases are not cinematic. They are inbox, calendar, ticket, CI, billing, and escalation workflows with clear wakeup rules.

7 min read · Updated 2026-05-27
Agent glyph behind a filter gate with a wakeup bell, noisy source cards, and fewer token counters after matching.
Agent economics

Your agent is not expensive. Your wakeup strategy is.

Token spikes in personal agents often come from broad wakeups, boot context, chat history, cron loops, and model depth applied to every branch.

8 min read · Updated 2026-05-27
Local runtime pulling matched work from a secure pending-delivery shelf fed by many event sources.
Local agents

Local agents need pull delivery

Webhooks work cleanly for hosted services. Local agents need durable pending deliveries they can pull, acknowledge, and resume.

7 min read · Updated 2026-05-26
MCP connector board turning many source tools into persistent future-condition watch cards.
MCP

MCP servers should leave future conditions behind

MCP gives agents tools for the present turn. Proactive systems also need durable watches that survive after the chat is over.

7 min read · Updated 2026-05-25
Cron clock tiles contrasted with a broad event-driven source mesh feeding a matched work queue.
Agent architecture

Cron is the wrong primitive for inbox and calendar watchers

Scheduled checks can ship a demo, but inbox and calendar assistants need event-aware wakeups with evidence and delivery semantics.

7 min read · Updated 2026-05-24
Token efficiency scene where many token blocks are filtered down to a few useful wakeup cards.
Agent benchmarks

Tokens per useful wakeup

A benchmark snapshot for WatchBench Email v0 and why proactive-agent systems should measure useful interruptions, not only task success.

8 min read · Updated 2026-05-26
Broad event-source console routing one high-value future condition into a work dispatch queue.
Agent infrastructure

What wakes the agent?

The most important question for proactive agents is not what model they use. It is which event is allowed to interrupt them.

7 min read · Updated 2026-05-22
OpenClaw, Hermes, Claude Code, and Codex-style agents around shared event infrastructure.
Agent infrastructure

What agent builders are learning from OpenClaw, Hermes, Claude Code, and Codex

The 2026 assistant stack is converging on the same pain points: context cost, handoff, permissions, local delivery, and reliable interrupts.

8 min read · Updated 2026-05-14
AI assistant receiving filtered app signals through Watchline's event-layer switchboard.
Agent infrastructure

The event layer is becoming agent infrastructure

Why proactive agents need an event layer between raw app streams and expensive reasoning turns.

7 min read · Updated 2026-05-14
OpenClaw-style red assistant receiving one matched interrupt through a local pull-delivery boundary.
Local agents

Local agents need interrupts, not bigger background loops

A broader view of why desktop and local agents need upstream event filtering before they can become truly useful.

6 min read · Updated 2026-05-14
AI assistant waking from one precise interrupt while scheduler and analyst agents wait nearby.
Proactive assistants

Proactive assistants need interrupt infrastructure

Why Pi, OpenClaw, Hermes, Claude Code, and Codex point toward event-driven assistant runtimes.

7 min read · Updated 2026-05-14
AI assistant beside cascade filter trays that reduce raw events before model reasoning.
Agent architecture

The cascade architecture for proactive agents

Why agent systems should filter events through cheap deterministic layers before spending large-model reasoning.

7 min read · Updated 2026-05-14
Codex-style cloud agent, local bot, and Claude Code-style mascot receiving routed events.
Delivery patterns

Agent events need delivery semantics

Why local agents, hosted agents, and MCP clients need different event delivery contracts.

6 min read · Updated 2026-05-14