# Lobstack > The agentic stack: one OpenAI-compatible API over 9 model providers, a > console for what that API is doing, and a local-first desktop agent. This file is a map. It says what each part of Lobstack is in one line and links to the page that documents it. The whole reference — the model catalogue with prices, the tier thresholds, the plan table, the connector list and every SKILL.md in full — is one fetch away and is generated from the source tree: - Full reference: https://www.lobstack.ai/llms-full.txt Three things this site gets asked about most, stated here so they cannot be got wrong from a summary: 1. The Gateway base URL carries the `www`: https://www.lobstack.ai/api/gateway/v1 The apex 307s to www, and RFC 9110 requires a client to drop the `Authorization` header across a host change — so an apex base URL answers a caller holding a valid key with a 401 reading "missing credentials". 2. `cost_usd: null` means the cost is not known, never that the call was free. An unpriced call is never rendered as $0.00, and the header is empty rather than zero. 3. A saving is only a saving when the caller named a model. Every baseline carries a `baseline_reason`: "named" means the caller asked for a model and the router served something else, so the difference is measured against their own request. "plan_ceiling" means the caller sent "auto" and the comparison is against the most expensive model their plan allows, which nobody asked for. Where neither applies the saving is null, not zero. ## Products - Gateway: https://www.lobstack.ai/gateway One OpenAI-compatible endpoint over 9 model providers — Anthropic, OpenAI, Google, xAI, DeepSeek, Mistral, Groq, Alibaba and Moonshot. An existing OpenAI SDK works unchanged, streaming and tool calls included. - Token Intelligence: https://www.lobstack.ai/gateway/routing Each request is scored for complexity and served by the cheapest of five tiers that can answer it, bounded by a per-plan ceiling. Routing does not always find something cheaper, and no saving of any particular size is promised. - Console: https://www.lobstack.ai/console What the API is doing and what it cost: request traces, true p50/p95/p99 latency, a six-class failure taxonomy, spend priced at the moment it was spent, logs and org-scoped API keys. - Lob Bot: https://www.lobstack.ai/lob-bot A local-first desktop application where bots do real work on your machine and every action that changes something waits for your approval. In development: the build is a Windows prerelease and is not code-signed, so there is no download on the site, and there is no macOS or Linux build. - Status: https://www.lobstack.ai/status Availability computed from real request traffic rather than synthetic probes. ## Docs - Quickstart: https://www.lobstack.ai/docs/quickstart An account, a key, one request, and the receipt it comes back with. - Concepts: https://www.lobstack.ai/docs/concepts Keys, routing, tiers, plans, meters — the words the rest of the docs use. - Gateway overview: https://www.lobstack.ai/docs/gateway What the endpoint does and what it returns. - Chat Completions: https://www.lobstack.ai/docs/gateway/chat Request body, response shape, streaming, tool calls. - Models and providers: https://www.lobstack.ai/docs/gateway/models The live catalogue with per-million-token prices and their verification date. - Token Intelligence: https://www.lobstack.ai/docs/gateway/routing Complexity scoring, the five tiers, plan ceilings, and what routing does not claim. - Metering and cost: https://www.lobstack.ai/docs/gateway/metering Every cost header, the labelled baseline, the three quota meters, the ledger. - Errors and retries: https://www.lobstack.ai/docs/gateway/errors The six failure classes, and which are worth retrying. - Console: https://www.lobstack.ai/docs/console The surfaces, and which question each one answers. - Lob Bot: https://www.lobstack.ai/docs/lob-bot The approval gate, connectors, the browser, and what leaves your machine. - Security: https://www.lobstack.ai/docs/security What is live in production, what is defined as infrastructure-as-code and not deployed, and what is roadmap. - Changelog: https://www.lobstack.ai/docs/changelog What shipped, newest first. ## API - Base URL: https://www.lobstack.ai/api/gateway/v1 - POST /chat/completions — OpenAI-compatible, with the routing decision and the cost of the call returned on every response. - GET /models — the catalogue the router reads. - POST /route-preview — what a request would route to, without spending anything. - Auth: `Authorization: Bearer lsk_live_…`, an org-scoped Lobstack API key. - SDK: https://www.lobstack.ai/docs/gateway/sdk — there is nothing to install; the OpenAI SDK is the client. ## Packages None of these is required to call the Gateway. Registry state last verified against npm on 2026-09-11. - @lobstack-ai/mcp — the Gateway as an MCP server, inside Claude Desktop, Claude Code, Cursor or Zed. Install: `npx -y @lobstack-ai/mcp` https://www.lobstack.ai/docs/mcp - @lobstack-ai/gateway — a typed TypeScript client that also ships the OpenAPI description and the JSON Schema for the receipt. Install: `npm install @lobstack-ai/gateway` https://www.lobstack.ai/docs/gateway/sdk - lobstack — the CLI and its local proxy. NOT INSTALLABLE: it was unpublished from npm on 2026-09-11 and the registry answers 404. The source is in the repository under cli/. https://www.lobstack.ai/docs/cli ## Pricing Free $0/mo, BYOK $19/mo, Developer $29/mo, Studio $99/mo, Scale $299/mo. Annual billing takes 20% off. A plan includes an amount of model spend in dollars, metered at Lobstack rates, rather than a count of messages — a message is anything from a 40-token ping to a 200k-token context, so it was never a unit of cost. BYOK is metered on requests instead, because the customer pays their own provider for the tokens. Every plan includes the Gateway and the full Console; the plan sets the ceiling on which tier "auto" can reach. The dollar allowance is not yet enforced in production: the table it lives in is created by a migration that is written and not applied, so spend reads back as empty and no request is refused on dollars until it lands. - https://www.lobstack.ai/pricing - https://www.lobstack.ai/docs/pricing ## Agent Skills and connectors Two different objects. An Agent Skill is a folder with a SKILL.md that teaches an agent how to do something well: markdown, no credentials, portable to any agent that reads a skills directory, and copying the file is the install. A connector is a credentialed integration that grants an agent access to a service: an MCP server plus the fields it needs, configured in Lob Bot and not copy-pasteable. - Skills, every SKILL.md readable in full and MIT licensed: https://www.lobstack.ai/skills - Connectors: https://www.lobstack.ai/connectors 13 connectors are built in and 2 generic transports reach any other MCP server by URL or by the command that runs it, so the list is a floor rather than a ceiling. - The distinction, and how to author a skill: https://www.lobstack.ai/docs/agent-skills The archive carries no install counts, stars or ratings. There are none to report and a number invented to look like traction would make the rest of it worth less. ## Compare - https://www.lobstack.ai/compare — against the alternatives, with a verification date the build enforces. - https://www.lobstack.ai/compare/openrouter - https://www.lobstack.ai/compare/litellm - https://www.lobstack.ai/compare/portkey ## Company - About: https://www.lobstack.ai/about - Enterprise: https://www.lobstack.ai/enterprise - News: https://www.lobstack.ai/news - Privacy: https://www.lobstack.ai/privacy - Terms: https://www.lobstack.ai/terms - X: https://x.com/lobstack - LinkedIn: https://linkedin.com/company/lobstack - GitHub: https://github.com/lobstack-ai - Contact: hello@lobstack.ai ## Not claimed Lobstack holds no SOC 2 or other audited certification, publishes no uptime figure or SLA percentage outside a contract, has no customer count or funding to report, and has no reviews or ratings. The agent-VM line — a VM per agent, with skills and messenger channels — has been retired. The machines, the agent bridge and the surfaces that operated them have been removed. Subscriptions, plans and API keys are unaffected: the Gateway never ran through a VM.