The agentic product OS
for your business

One workspace where AI agents and people run a product business from the same approved source. The AI drafts specs, sites, backends, documents, and replies as proposals or previews, and a person approves what ships. Project tools for building, selling, analysing, and teaching, and workspace tools for chat, time, invoicing, signatures, stock, and domains all read that source. Coding agents connect over MCP.

BYOK is a pilot for invited paid teams. A team owner or admin connects an OpenRouter key and chooses separate normal and Smart models. OpenRouter bills the team directly; these requests consume no CrystalSpec credits. Your subscription still applies. A failed or paused connection does not silently switch to paid CrystalSpec credits. Contact support@crystalspec.com about pilot access.

Not a ticket tracker first. CrystalSpec defines the product — flows, roles, data models, decisions, and test cases — and runs the work around it: spec-linked tasks and runs, strategy, sales, landing sites, shops, forms, backends, and onboarding all read the same published revision, while workspace tools handle chat, time, invoices, signatures, inventory, and domains. Teams with an existing tracker can push atomic tasks to GitHub, GitLab, Linear, and ClickUp instead.

No. The assistant only produces proposals. Each is validated, waits for your approval, and is recorded. Rejected proposals stay visible.

One Free workspace per verified account receives 500 AI credits each month, subject to Free model and usage limits. Creating, deleting, or transferring workspaces does not reset the allowance. Paid teams get 500 credits per billed seat monthly, in one shared pool. 500 credits cover $5 of AI usage at CrystalSpec rates; credits are not cash. Included credits reset monthly and do not accumulate. Paid teams can buy $10 top-ups for 1,000 credits, valid for 180 days. Reviewing proposals is free. On Free, wait for the monthly refresh or upgrade when your credits run out.

Text usage is charged at $25 per million tokens for Light and $60 per million tokens for Smart. Input and output tokens use the same rate within each model class. Credits are deducted from your shared balance at these rates. Web search and media have separate usage charges; reviewing proposals is free.

Team is $5 per member monthly or $60 per member billed annually, with no annual discount. Both plans include 5 GB of uploads per workspace. Paid workspaces get 1 GB more for each billed seat above five. Upload storage is a capacity limit, separate from project hosting transfer; contact support if you need more storage.

One draft at a time. The first editor locks the draft; everyone else sees the published version. Publish, then the next change starts from there.

Yes. Hosted MCP for Claude Code, Claude Desktop, and Cursor; GraphQL API; signed webhooks on publish. Spec exports, including PDF and DOCX, require a paid plan.

No. Each verified account gets one permanent Free workspace: unlimited projects, 1 member, and 500 AI credits each month. There is no trial or expiry. Team is optional at $5 per member per month, including the owner. Adding a second member requires Team: two members cost $10/month.

Free uses a low-cost model route, limits each provider response to 8,192 output tokens, and does not include Smart AI or web search. A separate monthly fair-use budget can pause Free AI before all 500 credits are spent. Every provider attempt, including retries, must fit within the remaining spending budget before it starts. Large requests can require more available budget than their final charge. Upgrade for paid-only AI tools.

Team unlocks additional teammates, Light and Smart AI, repository AI and document-maintenance automation, spec exports, and credit top-ups. Both plans include unlimited projects. Landing, Shop, and Backend are separate $10-per-project monthly upgrades available on either plan. Managed AI uses no-training provider routing on both plans. Your existing work stays in the same workspace when you upgrade.

No. CrystalSpec does not train models on your projects. Managed AI on Free and Team uses provider routes that exclude training on your prompts and project context. Retention and abuse monitoring follow provider terms. If you bring your own key, your provider terms apply.