🤖 AI agents

An agent that pauses days for a human yes/no, survives deploys, and runs each action once, with Workflow SDK.

A tool-using agent on AI SDK 7 with Zod tools, a remote MCP server, OpenTelemetry traces and an eval gate.

Connects what you've built (a voice guide, a morning brief, a chart, a scheduled loop) into one AI teammate with a written job description, the tools it may use, the guardrails it must keep and a human gate before anything goes out, then runs one cycle. Use when someone wants to go from separate prompts to an agent that does a recurring job, in ChatGPT, Claude, Gemini or any app with saved instructions.

Appends two or three sharp self-check questions after substantive advice, plans, estimates, or drafts so the reader pressure-tests the work instead of rubber-stamping it. Use once per conversation, after delivering a recommendation, estimate, plan, or draft someone will act on.

Decides whether a repeating agent task deserves a loop, then designs or reviews it so it can't spin, game its own check, or run a wrong answer to the end, with a yes-or-no done-check, "must not" boundaries, an independent judge, a retry cap and a staged rollout. Use when designing, writing or reviewing an agent loop (write-test-fix, a nightly fixer, watch until CI is green), or before putting an agent task on a schedule.

A tool that renders an interactive view (chart, form, table) inside chat clients via the MCP Apps extension.

House conventions for designing MCP servers and tools — tenant-from-credential auth, instructive errors, tiered access, and tool descriptions written for the calling model. Use when building, extending, or reviewing any MCP server or its tool definitions.

For builders. Creates a small local tool that answers questions about a CSV export from your ERP or tracking system without being able to change it (read-only, no network), with ten test cases it must pass, a log of every question, and a one-switch way to turn it off. It can optionally expose the same read-only tools to an AI agent over MCP. Use when someone with a coding agent wants to give AI safe access to business data and test it before trusting it.

A browser and phone voice agent with LiveKit Agents: streaming STT-LLM-TTS, turn detection, and tests.

A stateless MCP server on the 2026-07-28 spec with OAuth 2.1 sign-in, audience checks, and per-tool scopes.

Maps one business process step by step and marks every step green (AI may draft), amber (AI may check, a person decides) or red (a person only), with the reason, then turns the map into a one-page AI use policy for that process. Use when an owner, manager or compliance lead wants to use AI safely in a process with legal, customer or chain-of-custody stakes, or needs an AI policy their auditor would accept.

Holds non-code work that ships (client deliverables, newsletters, published pages, reports, customer-facing copy) until two independent reviewers, using the same pass/fail rubric and no shared context, both pass it; only flagged issues get fixed, fresh reviewers check each round, and after three rounds a person takes over. Use before publishing or sending anything where an invented fact, a missed requirement or a compliance slip would reach a client or the public, or to check a large generated batch.

Connects any agent or client workspace to the uplift.page prompt platform — the one MCP URL, signing in through uplift.page when a workspace tool asks, workspace keys for automation, and the plain-HTTP fallback. Use when setting up uplift.page access, onboarding a client workspace, wiring the uplift MCP server into an agent, or when a workspace tool reports a missing or invalid credential.