v0.1.3 · MIT · Open source

Pay for the AI you actually need

PromptBus routes Claude Code requests to the cheapest capable model. Simple questions use Haiku. Complex work stays on Sonnet. Savings add up without you noticing.

Get Started Documentation

terminal

$ npm install -g @promptbus/promptbus

$ promptbus install
$ ✓ Claude Code configured

$ promptbus start
$ ✓ Proxy on :4701 · Dashboard :4702

Every query costs the same — even the trivial ones

Claude Code sends everything to claude-sonnet-5. A one-line question costs the same as a full refactor. PromptBus changes that.

read_explain

"What does this function do?"

45 tokens, 0 files

$1.00per 1M input tokens

vs $$2.00 with Sonnet

small_edit

"Add input validation to createUser"

120 tokens, 1 file

$3.00per 1M input tokens

multi_file_refactor

"Extract auth into a service module"

340 tokens, 4 files

$3.00per 1M input tokens

Quick start

Install and start saving in under a minute

No account. No API key. No configuration. Just install, enable, and go.

Install globally

npm install -g @promptbus/promptbus

Enable for Claude Code

promptbus install

Start the proxy

promptbus start

View full docs

terminal

$ npm i -g @promptbus/promptbus
  + @promptbus/promptbus@0.1.3

$ promptbus install
$ → Add ANTHROPIC_BASE_URL to settings?
$   ✓ settings.json updated

$ promptbus start
$   ✓ Proxy listening on 127.0.0.1:4701
$   ✓ Dashboard at http://localhost:4702

$ promptbus status
$   ● Proxy is running

Smart defaults

Routes that make sense out of the box

Every task type is pre-configured with a sensible default. Tweak them anytime in config/rules.yaml or the dashboard.

Task type	Default route
`read_explain`	`claude-haiku-4-5`
`small_edit`	`claude-sonnet-5`
`test_generation`	`claude-sonnet-5`
`debug_loop`	`claude-sonnet-5 (never downgrade)`
`planning`	`claude-sonnet-5 (never downgrade)`
`multi_file_refactor`	`claude-sonnet-5 (never downgrade)`
`unknown`	`claude-sonnet-5 (fail safe)`

Safety-first: requests below the confidence threshold pass through without downgrading.

Models

Supports 16 Claude models

Route each task type to the most cost-effective model. Every major Claude variant is supported out of the box.

Model	Input / 1M tokens	Output / 1M tokens	Tier
claude-fable-5	$10	$50	Premium
claude-mythos-5	$8	$40	Premium
claude-mythos-preview	$8	$40	Premium
claude-opus-4-8	$5	$25	High
claude-opus-4-7	$5	$25	High
claude-opus-4-6	$5	$25	High
claude-opus-4-5	$5	$25	High
claude-opus-4-5-20251101	$5	$25	High
claude-opus-4-1	$5	$25	High
claude-opus-4-1-20250805	$5	$25	High
claude-sonnet-5	$3	$15	Standard
claude-sonnet-4-6	$3	$15	Standard
claude-sonnet-4-5	$3	$15	Standard
claude-sonnet-4-5-20250929	$3	$15	Standard
claude-haiku-4-5	$1	$5	Economy
claude-haiku-4-5-20251001	$1	$5	Economy

Dashboard

Real-time insights

Monitor routing decisions, track cost savings, edit rules, and export logs from a clean web interface.

PromptBus

Port 4701Model sonnet-5Floor 60%

Live

Active

1,247

Total Requests

843

Downgraded

67%

Reroute Rate

requests downgraded to cheaper model

Routes

$12.42

Cost Saved

$18.53

Original Cost

what you would have paid

$6.11

Actual Cost

what you actually paid

Requests6 requests

Time	Type	Requested	Routed To	Effort	Action	Latency
12:34:56	read explain	sonnet-5	haiku-4-5	Low	↓ Rerouted	1.2s	↗
12:35:12	small edit	sonnet-5	sonnet-5	Medium	→ Passthrough	3.4s	↗
12:35:45	debug loop	sonnet-5	sonnet-5	High	→ Passthrough	8.1s	↗
12:36:20	test generation	sonnet-5	sonnet-5	Medium	→ Passthrough	5.7s	↗
12:36:58	read explain	sonnet-5	haiku-4-5	Low	↓ Rerouted	0.9s	↗
12:37:30	multi file refactor	sonnet-5	sonnet-5	High	→ Passthrough	12.3s	↗

+ View all 1,247 requests in the full dashboard

Routing Rules

read_explain→claude-haiku-4-5low

small_edit→claude-sonnet-5medium

test_generation→claude-sonnet-5medium

multi_file_refactorNever downgrade

planningNever downgrade

debug_loopNever downgrade

unknownNever downgrade

Add rule

Advanced Settings

Confidence Floor

60%

Log Retention

days

Log Request Bodies

Codes/prompts may contain sensitive data

Save Settings

Export Logs

Start Date

End Date

Export CSV

Dashboard runs at http://localhost:4702

Frequently asked questions

Start saving on every API call

Install PromptBus in under a minute. No account. No config.

Get Started CLI Reference

MIT License