typed: A Claude Code-Compatible Backend With No 5-Hour Limits

Today we're launching typed - an AI CLI for coding. More volume per dollar, cheaper extra usage, and monthly billing instead of 5-hour reset windows. One curl installs the launcher; then run t (or typed) from any project directory. Works with Claude Code, Cursor, Cline, and any Anthropic-API-compatible client.

typed runs a different model under the hood. Most coding workflows feel identical; the migration page calls out where they don't. (Update, May 31: we benchmarked it -- same Claude Code, two backends, every task graded by a real test suite. typed and Opus 4.8 tied 14/14.)

Try typed from $10/mo →

The install

The fastest path is the typed launcher. One curl installs it with your API key wired in:

curl -fsSL https://app.typed.cloud/install.sh | bash -s <your typed key>

The dashboard pre-fills the command with your key plus a Copy button. Once installed, typed and t are on your PATH; both spawn Claude Code against typed.cloud with every env var set. typed status, typed disable, and typed enable let you bounce between typed routing and your Claude subscription without touching shell rc files.

Since launch, typed also ships its own first-party coding CLI, built in-house alongside the backend. It is opt-in: typed tui on makes subsequent typed / t runs launch the typed CLI instead of Claude Code, and typed tui off puts it back. It reads the same project surfaces you already have -- CLAUDE.md, settings.json permission rules, skills, slash commands, agents, hooks, and .mcp.json stdio servers. Section 1c of the migration page covers what it does differently.

If you would rather wire it up by hand, set three environment variables in your client:

export ANTHROPIC_BASE_URL=https://api.typed.cloud
export ANTHROPIC_AUTH_TOKEN=<your typed key>
export ANTHROPIC_MODEL=typed-xhigh

The API is wire-compatible. We run our own day-to-day on typed's first-party CLI now, but Claude Code and the other Anthropic-API-compatible clients (Cursor, Cline, Roo Code, aider) all work the same way against the same endpoint. Any tool that reads ANTHROPIC_AUTH_TOKEN and accepts a base-URL + model override talks to typed.

The third line is required: typed uses its own model identifiers (typed-xhigh is the default; typed-max, typed-high, typed-medium, and typed-low are also available). The full per-tier breakdown lives on the migration page.

You can flip back to Claude at any time -- typed disable if you installed the launcher, or unset the variables if you wired it up by hand. typed and Claude can coexist on the same machine, in side-by-side panes, with no shared state between them.

A Starter tier exists at $10/month (roughly half of Claude Pro's monthly usage) for lighter workloads. Top-ups work on every paid tier.

What's the same

Most coding workflows: refactoring, debugging, code generation, code explanation, test writing.
Image input: screenshots, mockups, error pastes - multimodal works.
Context comparable to Claude: ~500K tokens default on every text tier (typed-max / xhigh / high / medium / low), auto-upgrading to a ~950K long-context backend when a request exceeds it -- no beta header, no tier change. Multimodal requests cap at ~500K.
MCP servers: any MCP config you already use with Claude Code works against typed.
Prompt caching is on by default on every paid plan. Repeated-prefix workloads (system prompt, tool definitions, CLAUDE.md) get cache discounts automatically - you don't opt in. Caching engages most reliably on the default typed-xhigh tier; other model tiers cache on a best-effort basis.

What's different - be honest

Different model = different responses on edge cases. Most coding work feels identical; some won't -- on test-graded tasks they tied 14/14. The migration page calls out where it differs.
No Claude artifacts. typed doesn't render code previews in a browser tab. If you rely on artifacts heavily, keep Claude for that surface.
Coding-only product surface. typed is tuned end-to-end for coding workflows. General chat and creative writing will work but aren't the design target. Use Claude (or another general-purpose assistant) for those.
All sales final. Refunds are discretionary for billing bugs and confirmed service outages, not a marketing window. The full policy is published on the migration page and in the Terms of Service.

The full honest comparison lives at typed.cloud/migrate. If you're evaluating, read it first.

The math, for one user

Take a Claude Pro user on Sonnet (Claude's default coding tier) who consistently spends an extra $40/month on Extra Usage to finish the work the tier doesn't cover.

	Claude Pro + Extra Usage	typed Pro + top-ups
Subscription	$20	$20
Overage spend	$40 (Sonnet rates: $3/M input, $15/M output)	one or two $5-$20 top-ups against the monthly envelope
Total	~$60/mo	$20 + one or two $5-$20 top-ups
Reset window	5h rolling + weekly	monthly

For a Max-tier user paying Opus rates on Extra Usage, the gap is much bigger -- a typed top-up costs the same whether the work would otherwise have billed at Sonnet rates or Opus rates. You buy a $5, $10, $20, or $50 top-up against the monthly envelope; the price of that top-up does not change with the model you'd otherwise be paying for.

What typed is not

typed is not a Claude proxy. We don't route requests to Anthropic. typed runs on independent inference providers.
typed is not a free tier. We don't have one and we're not planning one. Customers arriving from Claude have already validated they'll pay for AI coding.
typed is not always-on. Our SLO is 99.9% monthly uptime with a live status page at typed.cloud/status. Past quota you get a 429 with a one-click top-up prompt (or auto-top-up if you've enabled it) - no hard cap, but no silent uninterrupted service either.

Privacy and data

Zero data retention is configured at the routing layer. Prompts and completions are not retained by upstream providers under our account configuration.
Our privacy policy discloses the cross-border data-flow chain in full, including the third-party providers we route through.
We don't train on customer prompts and we don't permit upstream providers to do so under our contractual terms.

If you're evaluating typed for a privacy-sensitive workload, read the privacy policy first. It's the disclosure surface.

How to try it

Sign up at app.typed.cloud with a magic-link email.
Pick a tier. Starter $10, Pro $20, Max $100 monthly. Annual at 10x monthly with two months free.
Generate an API key.
Run the one-line installer from the dashboard (with your key pre-filled), or set ANTHROPIC_BASE_URL=https://api.typed.cloud, ANTHROPIC_AUTH_TOKEN=<your typed key>, and ANTHROPIC_MODEL=typed-xhigh in your client by hand.
Run a real session against your codebase. Decide.

If you decide typed isn't for you, swap the variables back. You're never locked in.

Why we built this

Yaw Labs builds developer tools. We've been Claude Code users since launch. Two things consistently came up across our own usage and from customers on the Yaw Terminal community:

The 5-hour reset windows are friction during heads-down sessions.
Extra Usage adds up faster than people expect, especially at Sonnet ($3/M input, $15/M output) and Opus ($5/M input, $25/M output) rates.

typed is what we wanted as Claude customers: a coding-first AI service that doesn't punish heavy use.

If you've felt either of those problems, give typed an afternoon. Install with one curl, and you can evaluate it on the Starter tier for $10. Keep Claude as your primary; reach for typed when the windows or overage bite.

Update 2026-05-13: Team plans are live

Team and Team Max ship today for orgs that want shared billing and a hosted codebase index. Typed Team is $25/seat/month and carries the Pro-equivalent quota envelope per seat; Typed Team Max is $100/seat/month and carries the Max-equivalent quota envelope per seat. Both are monthly only for V1 - per-seat annual is deliberately deferred and we will revisit if customers ask. The whole subscription bills at per_seat_price * seat_count under our payment processor's volume pricing scheme.

The new surface Team and Team Max add is hosted codebase indexing. Local context works everywhere, on every tier and every Anthropic-API-compatible client; hosted indexing is the additional convenience for orgs that want a shared, server-side index across seats. It is not a missing feature on solo tiers. Storage and processing stay inside our us-west-2 AWS region, access is org-scoped, and codebases retain a 30-day grace post-cancellation before hard-delete.

Current pricing for all five tiers lives at typed.cloud/pricing.

AI CLI with more volume per dollar and cheaper extra usage.

One curl installs the launcher. Top-ups you control, priced the same whatever Claude model you'd otherwise be paying for. Monthly billing, no 5-hour windows. Magic-link signup, no lock-in. Team seats start at $25/seat/month with shared billing and a hosted codebase index -- see typed.cloud/pricing.