General Input

General Input's unit of work is a "specialist": a named agent that owns one job and holds the workflows, apps, skills and shared files behind it. You describe the process, and the platform writes it as clear, readable steps you can inspect before anything runs rather than a node graph.

Workflows come in two shapes, and the distinction matters for cost and reliability. A code workflow is a single main.ts that runs server-side once per trigger. An agent workflow spawns an LLM agent that reads an instructions.md procedure and calls APIs through bash, with a wall-clock cap and a reasoning effort set in agent.json. The vendor's own guidance is to reach for the agent only when "the exact steps aren't known in advance". Triggers are webhook, cron and poll, and an optional filter.ts can gate a noisy source before the body runs. Apps are React single-page applications in a sandboxed iframe that call named server handlers, which is how results come back as dashboards.

There is no free plan. Pro is $30 per seat per month with 4,800 credits per seat, and Max is $100 with 16,000, both capped at 50 members, and the first month on either is 80% off with the full credit allowance. Enterprise is quote-only.

Credits are the real meter. Overage is $0.0069 per credit on both tiers, plan credits are 25% cheaper than pay-as-you-go top-ups, and subscription credits do not roll over month to month. The vendor publishes its own run costs: a deterministic code workflow costs under 1 credit per run, an agent workflow 10 to 50, and building a workflow 30 to 150. A team running a handful of agent workflows hourly will clear 4,800 credits well before the month ends, and nothing on the page models that. Payment is card-only through Stripe, and plans can be cancelled at any time with the change taking effect next cycle.

Permissions are set per connection and then per action: the vendor's example allows Gmail reads and draft creation while blocking send and delete. A standing workspace policy is inherited by every member and their agents, including new joiners, and the vendor states that enforcement runs in the sandbox, beneath the credential, so a blocked call never leaves.

Credentials are encrypted at rest with AES-256-GCM and keys held in a separate secrets manager, injected at runtime and stripped before anything reaches the model. Every permitted call is logged with context, and the REST API's vault endpoints note that issuing a credential token is "logged with your reason". For a platform whose selling point is agents acting with live business credentials, that is the right place to put the controls.

The developer page names the geni CLI on npm and a per-workspace MCP server over Streamable HTTP at cloud.generalinput.com/v1/mcp/{organizationId}, authorized with OAuth 2.1, PKCE and dynamic client registration. Both discovery documents are published, RFC 8414 and RFC 9728.

Behind them sits a /v1 REST API with bearer auth and two kinds of key, personal keys that act as you and platform keys that act as the workspace. It covers the integration catalog, the credential vault, workflow and skill authoring, executions and traces, projects, memory, and six first-party hosted services including web search, page scraping and chat completion. Keys are issued from the dashboard without a sales call and can be revoked. The site also ships markdown twins of its pages, an llms.txt index and a real 404 body pointing at them, which is unusual and genuinely useful.

The model catalog spans Anthropic, OpenAI, Google, xAI and open-weights labs with per-million-token prices listed, models are switchable per task or conversation, an existing Claude, ChatGPT, Gemini, Grok or Muse subscription can be connected, and any OpenAI-compatible endpoint can be added with the buyer paying their provider directly.

The legal entity is Parla FM, Inc., doing business as General Input, governed by Delaware law, and the about page is a single founder letter signed by Jack. No funding, headcount or founding date is stated. The subprocessor list is unusually complete: Google Cloud, Cloudflare, Vercel, OpenAI, Anthropic, Vertex AI, xAI, Fireworks, OpenRouter, AssemblyAI, E2B, Stripe, Postmark, Sentry, Datadog and more. No SOC 2 report, ISO certificate or DPA is published, only CASA verification on the plan table, and the terms disclaim any uptime warranty.

A buyer who needs a free tier to evaluate will not find one, and the 80%-off first month is a discount rather than a trial. SSO, white-label domains, phone support and on-premise deployment are all Enterprise-only, so a 20-person team that needs identity-provider sign-in is talking to sales whatever the published seat price says.

The pricing page states the Business plan tops out at 50 members while also offering it for teams of any size, and the two marketing figures for integration breadth do not agree: the home page and plan table say 850+ tools, while the Agent Firewall FAQ says rules apply "across 80,000+ tools". The larger figure most likely counts individual operations rather than services, but the site never says so. Nothing on the site names a region option for hosted data, which is stored in the United States.