Engineering service
Embedded AI & browser assistants
Chrome extensions and sidepanel experiences with in-browser RAG and real actions via tool gateways — not another detached chat tab.
- Surface
- Plasmo · Chrome extension
- Privacy
- IndexedDB · scoped tools
- Compute
- Lambda · Workers
In depth
The best assistant UX meets users in the tab they already use. I build extension-first experiences — Plasmo, React, TypeScript — with page and selection context, tool execution at the edge, and privacy-aware storage when needed.
At Get Magic, that meant the Magic Assistant and open-floor assistant patterns: in-browser RAG with IndexedDB for sensitive cases, plus MCP-style tool servers for CRM, calendar, email, and internal APIs.
What you get
Extension UX: context, permissions, and safe tool surfacing
On-device or hybrid retrieval patterns for playbooks and history
Tool gateways with least privilege and audit trails
Low-latency execution paths (edge + server) for real actions
Prompt-safety and injection-hardening patterns for tool-enabled flows
How we work
Anchor on the workflow
Which pages, selections, and actions define success?
Prototype the extension shell
Context capture, auth, and minimal RAG loop.
Wire tools carefully
Scoped adapters instead of “give the model everything.”
Enterprise gates
Feature flags, rollout controls, and sidepanel UX polish.
Examples & past work
Outcomes you can expect
- Ship Plasmo extensions that respect page and selection context for assistant workflows.
- Use on-device stores (e.g. IndexedDB RAG) when sensitive context should stay local.
- Connect assistants to CRM, calendar, email, and internal APIs through scoped MCP-style adapters.
Questions, answered
Why Plasmo?
It accelerates MV → production for MV3 extensions with sane React tooling — the same patterns port if you standardise on a different scaffold.
How do you handle sensitive client data?
Data minimisation, on-device stores when possible, scoped tool access, and clear retention boundaries — aligned to how Magic approached high-trust assistant contexts.
Can assistants take actions in our SaaS product too?
Yes. Extensions and in-product panels can share the same tool contracts; the UX changes but the safety model should not.
Book a free 30-minute discovery call to investigate your work and needs
I will map constraints, risks, and a practical first milestone — whether that is agents, retrieval, ingestion, extensions, or full-stack SaaS delivery.
Other services
Multi-agent workflows with planning, memory, typed tool calls, and human checkpoints — from creative-strategy platforms to assistant co-pilots.
Vector + relational retrieval, semantic chunking, and re-ranking so LLM outputs stay grounded when catalogues and documents get large.
Schedulers, retries, and normalised pipelines from ads APIs, social platforms, and internal services — built for scale and observability.