Systems
Agentic systems
Reviewable multi-agent workflows, browser automation, observability, escalation paths, and boring fallbacks that keep the system legible under stress.
Agency / execution layer
Polygonface Agency
Polygonface builds websites, operator tooling, and agentic systems for teams that need real delivery.
Polygonface is the execution layer behind the publication.
We design and ship websites, internal tools, agentic systems, and integration rails for teams that need the work to become real.
The agency is embedded into the larger publication ecosystem, but it is not the primary identity of the front page.
Agentic systemsOn-prem LLMConsultation + educationWeb development
CloudflareOpenAIAnthropicOllamavLLMAstroSvelteKitPostgresD1R2ListmonkMauticSlackNotionCloudflareOpenAIAnthropicOllamavLLMAstroSvelteKitPostgresD1R2ListmonkMauticSlackNotion
What Polygonface actually does
Clear work, not AI theatre.
Private stack
On-prem LLM
Local or private-model deployments, retrieval and policy boundaries, internal tooling surfaces, and constrained operator paths for sensitive environments.
Advisory
Consultation + education
AI operating models for public and private-sector teams, workshops, executive framing, internal playbooks, and training that lowers confusion instead of raising theatre.
Surfaces
Web development
Astro and SvelteKit frontends, editorial sites, control rooms, partner portals, and service surfaces where humans and agents can actually work.
Selected systems
Proof of shape, not moodboard screenshots.
Showcase
Websites are becoming operational surfaces.
Publishing, capture, review, automation, governance, and internal operator actions are increasingly meeting on the same frontend. We design around that reality from the start.
Editorial system
Dispatch + operator CMS for a newsroom-first site with a distinct agency layer.
Publishing, feeds, media variants, and operator review in one Cloudflare-native stack.
Astro frontWorker CMSD1 / R2 / Images
Operator tool
Presentation console for live rooms, fullscreen media, and public-safe playback.
Live deck playback, modal media handling, and a presentation console tied to the wider site stack.
Schema-gated decksLive media controlsPublic-safe shell
Private stack
Private LLM ops handoff for a team that needed policy, retrieval, and operator guardrails.
Policy boundaries, retrieval design, and operator surfaces for high-trust model deployments.
Private inferenceRetrieval lanesAudit-aware controls
Workflow rail
Reviewable browser workflows for teams tired of brittle automation theatre.
Visible state, retries, operator checkpoints, and automation that stays legible under stress.
Browser tasksRetry railsOperator review
Delivery model
Short cycles, visible decisions, train-the-team handoff.
Engagements typically begin with diagnosis and scoping, move into surface design and build, and end with training or governance material so the organization can actually run the thing.
Step 01
System diagnosis
Map the real queues, handoffs, ownership, and breakpoints before touching the frontend.
Step 02
Surface design
Decide what belongs on the public site, what belongs in the operator layer, and what should never be user-facing.
Step 03
Build + integrate
Ship the frontend, auth boundaries, storage, tracking, delivery pipeline, and content model as one coherent system.
Step 04
Train + operationalize
Hand over playbooks, internal education, monitoring, and enough structure that the thing survives contact with real staff.
Where this lands
Public and private-sector teams that need AI to become legible, ownable work.
- Public-sector teams who need AI guidance without procurement theatre.
- Private-sector operators replacing fragmented no-code stacks with something they can actually own.
- Leadership teams that need executive framing, staff education, and a sane path to implementation.
- Organizations with sensitive data, private infrastructure, or a need for on-prem model deployments.