Manhattan’s pedestrian pulse, reconstructed as a changing density field from public camera observations.

Today I’m announcing klosr, a spatial intelligence platform from aimez.ai. Its first pilot brings modeled shade and observed sidewalk occupancy into pedestrian route evaluation across Manhattan. An interactive feature demo is now live at klosr.aimez.ai, with early test access opening soon.

A route on a map becomes a walk once someone steps onto the street. Distance and travel time remain the default measures for comparing pedestrian routes. Neither measure expresses the shade or sidewalk occupancy encountered from block to block. klosr gives each street edge a condition cost before candidate paths are ranked. The recommendation can change as those conditions accumulate across the walk.

klosr brings street conditions into the route before the walk begins

Standard pedestrian navigation prioritizes distance or travel time alone, ignoring dynamic street conditions such as shade exposure and sidewalk crowding. Because a single block carries variable physical costs across the day, static distance metrics cannot represent human-scale walking choices. klosr addresses this limitation by projecting camera observations and reconstructed ground shade onto a walk-graph, producing interpretable comfort-efficiency trade-offs for pedestrian route choice.

Route evaluation from Grand Central to Carnegie Hall. Sidewalk crowding diverts the adaptive route around dense Midtown crosswalks.

What happens when shade and sidewalk crowding become part of the route?

Two routes can cover the same distance and expose a walker to different conditions. Shade moves across the day while sidewalk occupancy changes by place and time. A route model can compare physical exposure only after each condition has been assigned to the walk graph.

Research on pedestrian route choice describes a context-dependent process shaped by perceived environmental attributes. Recent comfort-aware routing work translates thermal exposure into graph costs by assigning temperature and canopy data to street-network edges.

klosr assigns modeled shade and observed sidewalk occupancy to the Manhattan walk graph. Edge costs accumulate along each candidate path, so local conditions can change which route ranks first.

Pedestrian routing forms the initial evaluation zone because every path passes through a sequence of local conditions. The Manhattan pilot tests how accumulated shade and sidewalk occupancy change route rankings.

Mapping street observations into physical walk-graph edge costs

Public NYCTMC camera frames provide local sidewalk occupancy observations. Pedestrian detections inside sidewalk and crosswalk regions produce a calibrated density signal for nearby walk-graph edges.

The edge-cost model combines the occupancy signal with reconstructed shade for each walk-graph edge. The routing engine sums edge costs across every candidate path within the distance constraint, then recommends the path with the lowest total. A change in one block’s conditions can alter the cost of an entire path and produce a different recommendation.

Benchmark route candidate scoring between Gansevoort Peninsula and Washington Square Park at 14:00 EDT. The shortest route measures 1,808 m with 53% length-weighted shade coverage, while the shade-seeking route extends the walk to 1,926 m and reaches 86% shade coverage, gaining 33 percentage points of shade for a 6.53% distance increase.

Reconstructed ground shade at Greenwich Street and Perry Street, 2:00 p.m. A 2-meter grid computes direct sunlight across the sidewalk and finds 204 of 504 squares shaded, 40.48% shade coverage.

Solar position determines the path of each ray. Building massing and canopy geometry determine whether the ray reaches the ground. The resulting shade values become exposure costs on nearby walk-graph edges, where each path accumulates the conditions encountered along its route.

By integrating calculable solar exposure with street-level pedestrian activity, the pipeline constructs dynamic edge costs across time. Camera-derived density fields provide the observed state for research evaluation, while Field-JEPA estimates the field 30 minutes later to test whether recent observations carry enough information to anticipate near-term crowding.

Observed pedestrian-density fields across three Monday morning holdout windows appear above their Field-JEPA 30-minute forecasts. The forecasts come from an offline research evaluation and do not drive the current routing demo.

In offline evaluation across 32 holdout windows spanning 339 Manhattan camera locations, Field-JEPA achieves a mean absolute error of 0.0901 in crowding rank at the 30-minute horizon, compared to 0.1050 for the persistence baseline. This represents a 14.2% reduction in forecast error, with a Spearman rank correlation of 0.725 between predicted and observed crowding.

Borough expansion and spatial transfer

The initial deployment covers pedestrian routing in Manhattan. Extending coverage across New York City will test whether the spatial pipeline transfers across new boroughs as local shade and occupancy datasets come online. Replicating the system in another city will require constructing and validating local street observations first. Vehicle routing will require a distinct condition model because driving constraints differ fundamentally from walking.

Storefront placement and site selection

Expanding spatial coverage increases the footprint over which block-level sidewalk occupancy can be compared. Upcoming commercial pilots will test a ranking model that matches campaign briefs against host-controlled storefront surfaces using local foot-traffic data. Recorded placement outcomes will link street conditions to campaign performance, creating an empirical record for future site selection. Property economics and leasing constraints will require separate financial models before site selection becomes a standalone product.

Take a klosr look

Early access registration for klosr is now open. Sign up for release updates and preview access as the Manhattan pilot rolls out. Commercial partners and research evaluators can indicate interest through the same form.

Request early access

In the first aimez post, I want to introduce wavves, a free Cursor plugin for moderator-led multi-agent work with executable checks and explicit thread handoffs.

Developers have been attacking agent work from several directions. Steve Yegge’s Gas Town treats agents as a fleet. A Mayor assigns work to persistent agents, while Beads and git-backed Hooks preserve state across sessions. pstack, by Lauren Tan, works from the execution side by giving individual Cursor agents engineering discipline, review habits and playbooks for parallel work. Orchestrate pushes fan-out into cloud-agent trees.

Together, those systems make agent work larger, faster or more disciplined. The remaining gap sits at the moderator layer. The operator-facing thread still needs a compact record of scope, locked decisions, accepting checks and pickup state, so the next thread can resume without replaying the chat. wavves addresses that gap by turning operator intent into an alignment packet before sub-orchestrators and parallel runners are deployed.

State capture for durable multi-agent alignment

Distributed sessions need two durable routes. Governance and orchestration need a route before work fans out, with scope, grounded facts, preserved decisions and file boundaries available to every runner. Integration, escalation and adoption need a route as work comes back, with pickup paths, check records, accepted results and handoffs available to the moderator and successor threads.

A chat transcript is a weak recovery surface because decisions, tool output and failed attempts are mixed into the same stream. Summaries can reduce volume while dropping the constraint that governs the next action. Launch records can remain trapped in a dead thread and output can appear ready before the accepting check is recorded beside it.

wavves moves those records out of the conversation through an alignment packet from the moderator, check records from runners and handoffs that carry active work into the next thread. Distributed work becomes easier to review because the shared record lives outside chat.

A walkthrough of the setup

wavves is a set of written contracts, plain markdown files under version control.

Every lane begins with a moderator pass that turns operator intent into alignment infrastructure. The packet records scope, grounded facts, preserved decisions, file boundaries and acceptance checks, giving the operator-facing thread a basis for dispatching sub-orchestrators and giving runners a record target before any result is accepted.

When the working context becomes large, the outgoing instance writes a handoff file that describes active work, accepted results and unresolved work. The successor hydrates from that file set rather than from accumulated conversation.

The operating rule moves captured state into alignment packets, check records and handoff files instead of leaving it in chat history.

Context size and model cost

Thread rotation changes the amount of context presented to a successor agent. The measurements reported here describe the mechanism and one observed rotation, while a controlled comparison of token consumption remains future work.

Why the design cares about tokens. In a long thread, every new turn re-sends everything already in the window. Tool output, findings, failed attempts and settled decisions all ride along, so the same accumulated context is billed repeatedly. Context accumulation also reduces retrieval reliability while increasing cost.

What wavves does about it.

During one rotation, a successor started from a standing contract and handoff file totaling about 16 KB instead of an archived transcript of roughly 1.5 MB. The comparison illustrates a bounded starting state rather than a measured token-saving percentage.

A scope packet compresses what a fresh agent needs into a few kilobytes of locked decisions instead of re-deriving them from conversation. A real lane packet here is on the order of 4 KB. The packet replaces repeated operator restatement when a new runner spins up.

Heavy tool output and parallel work live in background runners with their own windows, leaving the operator-facing thread to scope work and reconcile returns.

The current package also records recommended model tiers in the lane charter. The moderator can reserve high-reasoning models for architecture, integration, adversarial review and acceptance while assigning faster models to inventory, link checks and other mechanical scans. When the runtime exposes explicit model selection, the moderator can pass that setting during dispatch. When it does not, the recommendation remains in the dispatch record and the runner reports the enforcement gap.

What the literature says. Agent benchmarks now treat context management as a first-order cost problem rather than a minor engineering detail. Lindenbauer and colleagues (“The Complexity Trap,” 2025, arXiv:2508.21433) compare strategies on SWE-bench Verified and report that simple observation masking roughly halves cost relative to an unmanaged agent while matching solve rate. The same study reports that a hybrid of masking plus selective summarization beats pure summarization on cost by about 11%. JetBrains Research wrote up the same line of work in December 2025 (efficient context management). The context-management papers study masking old tool outputs inside a single thread. wavves uses a different mechanism by pushing state into files and hydrating a fresh thread from those files when the old window gets heavy. The shared goal is to move bulk out of the billable context.

Older work on long contexts (“Lost in the Middle,” Liu et al., 2023, arXiv:2307.03172) shows that models use information less reliably as context grows, especially material in the middle. Shorter, file-backed hydration can lower cost while making the relevant state easier for the model to use.

For workloads dominated by large file reads or long tool traces, workers still spend tokens doing the reading, while the alignment structure partitions expensive work, keeps the operator thread small, avoids re-ingesting an entire long thread of chat during rotation and makes model choice reviewable before dispatch. A measured savings claim would require actual token or billing records from the run. The whitepaper has more on the accounting if you want it.

Scope and limits

The observations come from one operator working across one family of repositories. A controlled comparison against unstructured runs and a team-scale deployment study remain future work. The records preserve the fact that defined verification ran, while the operator still owns the quality of the gate.

The whitepaper and the plugin release

The companion whitepaper documents the method in greater detail together with records from the runs summarized here and is published on this site as a subscriber post.

The system is packaged as a free plugin called wavves. The three skills set up the orchestrator home, define and dispatch wavesets and rotate orchestrators with assigned identities under an MIT license independent of any editor vendor. If you run agent work in Cursor, you can install it with one line.

The current release captures scope, dispatch records, acceptance checks, model-tier recommendations and handoff files. A cost estimator would still need actual token or billing records from the environment, but the charter now records runner count, expected file reads, expected context and recommended model tier before execution. The next release can extend the same framework with measured token accounting and automatic rotation of moderator threads at configurable thresholds. The planned additions build on the existing scope and handoff system rather than replacing it.

/add-plugin wavves