32.78°N · 79.93°W CHARLESTON · SOUTH CAROLINA family owned · est. 2023 ● taking Q3 '26

Agentic AI, built porchside.

Palmetto Interactive is a Charleston studio building agentic systems, memory infrastructure, and the developer tooling that makes AI actually ship — and we put the same engineering behind Lowcountry businesses whose sites, stores, and ads just need to work.

Start a project → See selected work ↓

Agents running

7live●

Tokens / min

1.43ktok/m

Tasks shipped · '26

312YTD

14:02:11 recall → membl-cloud/steward → conflict-resolver.md 14:02:14 capture → lantern/squad-dispatch pattern 14:02:17 eval → recall p50 12ms · p95 41ms · pass 14:02:20 recall → chalets-by-the-lake/booking → tool-registry.json 14:02:24 capture → membl-cli release notes drafted 14:02:28 recall → shelter-fitness/billing → stripe-webhooks.ts 14:02:31 graph → +47 nodes · +112 edges · 3 superseded 14:02:35 eval → agent-eval-bench@0.4.1 · 31 passing 14:02:39 recall → membl-io/savings-ledger → query-shape.sql 14:02:42 capture → oswald-cooke/listings → crm-sync.md 14:02:11 recall → membl-cloud/steward → conflict-resolver.md 14:02:14 capture → lantern/squad-dispatch pattern 14:02:17 eval → recall p50 12ms · p95 41ms · pass 14:02:20 recall → chalets-by-the-lake/booking → tool-registry.json 14:02:24 capture → membl-cli release notes drafted 14:02:28 recall → shelter-fitness/billing → stripe-webhooks.ts 14:02:31 graph → +47 nodes · +112 edges · 3 superseded 14:02:35 eval → agent-eval-bench@0.4.1 · 31 passing 14:02:39 recall → membl-io/savings-ledger → query-shape.sql 14:02:42 capture → oswald-cooke/listings → crm-sync.md

SECTION 01 / SELECTED WORK06 engagements2023 — 2026

Selected recent work.
Most of it ships under someone else's name.

We work at the studio scale — three to five engagements at a time, six months average, mostly under NDA. Click a row, or a node on the timeline, to see what we actually did.

XU '06

Startup → Acq.

Wall Street

Palmetto Interactive

2006201420232026 now

engagementmembl · flagship20+ yrs · founder → wall street → studio

2026● featuredmembldeveloper platform · in developmentRustTypeScriptSQLiteTemporal→

2025 — ongoingConfidential — e-commerceshopify replatform · paidShopifyMeta AdsGoogle AdsGA4→

2025Confidential — professional servicesAI enablementPythonOpenAIZapierNotion→

2024Chalets by the Lakeunified booking platformWebAutomationStripe→

2024Shelter Fitnesscustom applicationFull-stackPostgresStripe→

2023 — 2024Oswald Cookewebflow · crm · it strategyWebflowCRMCloudSecurity→

SECTION 02 / PLATFORM 04 products status · in active development

What we're building when nobody's billing.

The membl platform is our own bet — a developer suite for agentic engineering, built in the open from Charleston and dogfooded on every client engagement. Memory first, but far more than memory.

membl

public beta

Persistent, portable memory for developers and AI agents. A local-first Rust daemon turns your work into an evidence-backed knowledge graph — provenance on every fact, conflict resolution, ~12ms recall, and a savings ledger that shows the tokens you didn't burn.

· brew install membl
· knowledge graph · provenance · TTL
· Claude Code, Codex & Cursor hooks
· savings analytics

membl.io ↗

membl Cloud

in development

The paid tier: encrypted sync and backups, shared team memory, a hosted dashboard, and savings analytics — so your whole team's agents stay current across every machine.

· cloud sync + backups
· team workspaces · audit log
· hosted dashboard
· savings ledger

Lantern

in development

A self-contained Rust runtime that orchestrates squads of AI coding agents on your machine — MCP server, git-worktree isolation, durable Temporal-backed state, and recovery when an agent wanders off.

· multi-agent squads · MCP
· git worktree isolation
· durable workflow state
· one binary per machine

membl Studio

research preview

A spec-driven engineering operating system: intent goes in, parallel agent sessions run in collision-proof worktrees, and a human reviews instead of task-manages. CLI and desktop app share one registry.

· collision-proof sessions
· per-role worktrees + port ranges
· Tauri desktop + CLI
· provider-agnostic agents

SECTION 03 / CAPABILITIES 06 disciplines last updated 2026.06

We build the useful half of agentic AI —
the part most teams hand-wave past.

Five disciplines, one team, one zip code. Engagements typically pair two of these — strategy + product, or infra + advisory. We don't staff to a model; we staff to a delivery.

Agentic systems

Multi-agent orchestration with real tool use — file edits, shell, browsers, internal APIs. We design the planner, the controller, and the evals before we write the prompts.

· planning · routing · verification
· tool registries & sandboxes
· human-in-the-loop checkpoints
· long-running session memory

LLM infrastructure

Memory, retrieval, and evaluation that survives a real codebase. Hybrid stores, conflict resolution, recall traces — the unglamorous plumbing that makes models look smart.

· hybrid retrieval (BM25 + vector)
· memory graphs · provenance · TTL
· eval harnesses · regression tracking
· cost & latency budgeting

Custom AI products

Full-stack ship-it work. Backend in Rust or Python, typed APIs, a dashboard your team actually wants to open. We work in week-long iterations against a written contract of done.

· 0→1 MVPs · 6–12 week sprints
· Rust · TypeScript · Postgres · AWS
· observability · audit · billing wired in
· handoff docs an engineer can actually read

Developer tooling

CLIs, daemons, IDE plugins, and the quiet local-first surfaces engineers prefer. If it has a prompt, a port, and a config file, we have probably built one.

· CLIs · daemons · file-system glue
· IDE & editor extensions
· local-first · single-binary distribution
· install scripts that don't ask questions

Strategy & advisory

For non-AI-native teams making their first real bet. Two-week reviews, one-page roadmaps, eval frameworks you can run without us. We are happy to talk you out of it.

· two-week technical reviews
· build vs. buy · vendor pruning
· team coaching · pairing weeks
· roadmaps you can actually execute

Things we will not do

Generic chatbots. RAG-as-a-service. Sentiment dashboards. Decks with rocket emojis. If the work doesn't move a number you can defend in a board meeting, we are the wrong studio.

· no chat-on-everything
· no demos without evals
· no headcount-shaped engagements
· no NDAs that prevent learning

SECTION 04 / PROCESS 04 phases avg engagement 11.4 wks

How we work, in plain language.

A studio process, not a consultancy process. We staff to the work, ship every week, and write things down. If a phase isn't delivering, we say so on a Friday.

Phase 01

Read the room

Two-week paid review. Codebase, agent traces, eval gaps, the politics. We deliver a written plan even if you don't hire us next.

duration · ~ 2 weeks

Phase 02

Write the contract

A one-page contract of done — scope, evals, latency budget, the bill. Nothing in the engagement is allowed to be a surprise to either side.

duration · ~ 1 week

Phase 03

Ship in weeks

Week-long iterations against the contract. A shipped surface every Friday — even if it's a CLI flag — and weekly written notes you can forward.

duration · 6 — 16 weeks

Phase 04

Hand it back

Runbooks, evals you can run, the on-call we trust. We leave when your team is faster without us than with us. Then we answer the phone.

duration · ongoing

SECTION 05 / CONTACT accepting Q3 '26 avg reply < 24h

Start a project.

We take three new engagements a quarter. Tell us what you're trying to ship, and what's in the way. The form below is read by a human in Charleston.