For agents. By agents. Your data. Your machine.

Adopt AI agents reliably. Across every role, every department, every team. Built for agents and humans from day one.

For solo builders shipping a side project, teams replacing manual workflows, enterprises rolling out agents at scale — many primitives, in one stack that scales with you.

Standalone packages. Optional composition. Polyglot libraries (Go, TypeScript, Python, Rust, PHP). CLI apps in the best language for the job.

Protocols: MCP · A2A · ACP · AGNTCY · gRPC · REST · WebSocket · SSE · Webhook

Vendors: Anthropic, OpenAI, Google, Ollama (models); GitHub, GitLab, Bitbucket, Gitea (repos); Linear (PM) — built in. BYOA: Bring Your Own Adapter, every interface is yours to implement.

No lock-in.

Split across seven disciplines — not steps to finish.

_{* Affiliated, not hop-top org — ctxt (context.help), mxhook (mxhook.com), xray (xray.codes). † vein planned, not yet released.}

Every package is standalone. Adopt one. Adopt a row. Adopt the matrix. Mix with what you already use; replace what you outgrow.

Seven disciplines, seven incumbents. hop.top doesn't try to be all of them in one product. It gives you the piece, you keep the rest.

hop.top isn't an AI coding agent. Claude Code, Codex, Gemini CLI, OpenCode, Aider, Pi, Cline — those are the harness around your model. hop.top is the layer that manages them: verifiable identities (aps), real task workflows (tlc), call recording (xrr), evaluation (eva), tool integrations (x402, foo, pod, ...).

Whole-stack monoliths — OpenClaw, Hermes, PrintingPress — bundle opinions for one mode. hop.top gives you both: seven discrete packages you compose à la carte, and hop — the all-in-one CLI when you want the bundle. Compose. Or brew install hop && hop init. Your call.

One spec. Five runtimes. Parity enforced, not promised.

_{✓ full implementation · ◐ envelope I/O only at v0.1, full runtime planned ¹ stem's Go SDK is the reference runtime; other languages parse + serialize crtx envelopes so non-Go services can consume sessions produced by stem.}

Parity is contractual:

• Cross-language conformance tests ship in poly-*/tools/parity/. Drift breaks CI, not your code.
• Shared scenario fixtures mean every SDK answers the same inputs identically — verified by make test-parity.
• See the receipts: poly-cite/tools/parity, poly-aim/docs/sdk-parity.md.

Framework for building new agent-first CLIs: kit (Go). Use kit when you're starting a new agent-native CLI; use the libraries above when you're adding capability to an existing one.

Libraries — polyglot (Go, TypeScript, Python, Rust, PHP):

• cite — Custom URI schemes — parse, validate, register
• xrr — Multi-channel interaction recorder/replayer
• aim — models.dev registry client — agent-safe model lookup
• agntcy — Polyglot SDK suite for the AGNTCY Agent Directory Service
• c12n — LLM request classification — route to right model by signal
• stem — AI coding agent agnostic sessions

Framework — for building new agent-first CLIs:

• kit — Polyglot framework for building agent-friendly CLIs

CLI apps:

• aps — Local-first agent profile system — isolated profiles for commands and workflows
• ben — General-purpose benchmarking — "which approach is better, and by how much?"
• crm — Customer relationship CLI — vCard contacts, vJournal interactions, pluggable sync
• cxr — Domain-agnostic dispatch runtime — routes by capability/tool intersection
• eva — Behavioral contract enforcement on AI agent responses — declarative YAML specs
• fit — Train small advisor models to steer black-box LLMs without fine-tuning — polyglot serving + GRPO training, frontier model never modified
• foo — Pipe text through LLMs from the terminal — pattern-based prompts, streaming
• git-hop — Multi-branch parallel worktrees — isolated environments, deterministic ports
• gym — Universal package manager for agentskills.io skills
• ibr — AI-powered instruction parser — natural-language → Playwright actions
• inv — Invoicing-as-a-service — composer + lifecycle, multi-channel core, event-bus interop with fin
• pod — Session, model, and tooling layer on top of any remote compute
• rlz — Case-YAML → per-AI release pack generator (changelogs, notes, social)
• rsx — Repo signal extraction — trust, activity, dependency risk
• rux — Interactive Terminal Execution Runtime
• tlc — Multi-agent task orchestration with Task Line Syntax
• upgrade — Self-upgrade library for hop family CLIs — Go package + @hop/upgrade (ESM) + @hop/upgrade-ts
• vstar — Calendar/vCard-shaped data convention for agentic systems (RFC 5545 + 6350)
• wsm — Workspace state manager — mutation history, access control, handoffs
• x402 — Protocol-agnostic x402 payment module for agent-native wallets
• xat — Cross-Assistant Tester — cross-CLI conformance + regression harness for AI-assistant plugins (Claude Code, Gemini, Codex, OpenCode)
• hop — AI work toolkit

GitHub Actions:

• 12fc — Verifies CLI conformance to the 12-Factor AI-CLI spec

Affiliated — separate orgs, vanity domains:

• ctxt — Decentralized, local-first context engine (context.help)
• xray — Codebase mapping (xray.codes)
• mxhook — Inbound email webhooks (mxhook.com)

• License — Most packages MIT; some Apache 2.0; rlz under BSL-1.1. See each repo's LICENSE for specifics.
• Releases — release-please + Conventional Commits. Rolling major tags. Published to npm, PyPI, crates.io, Packagist, Go vanity.
• Cadence — Independent per-package. Each ships when ready; no linked-version coupling.
• Stability — Pre-1.0 packages carry visible warnings. v1+ packages follow semver strictly.
• Security — security@ideacrafters.com · SECURITY.md

hop.top → — full catalog, docs, install paths

Community: Contributing · Security · Code of Conduct

hop.top is sponsored by Idea Crafters & Experts AI.