CacheDeep, OpenAI-compatible DeepSeek Flash

約官方高峰 0.5×,快取命中打骨折。長時間任務更省錢。

Tokens
Usage
Completions
Caching

What is CacheDeep

CacheDeep relays DeepSeek Flash over an OpenAI-compatible /v1 API. One model name, listed cache-hit pricing, and a console for keys and balance.

Call deepseek-flash through an OpenAI-compatible /v1 endpoint. Text and image requests share one model name; routing stays on the server.

Too much billing noise pulling you away

from the product you’re shipping

(
)
Forgot which key was liveWhere's that file?How much did that prompt cost?Did cache hit on the prefix?Which group am I billed on?Reconciling logs. Again.Is this within quota?"Just one more change"Where's that client task?Chasing usage rowsDid I rotate that sk-?

That's why CacheDeep exists

That's why CacheDeep exists

Console

New key
New
Scope

What

Add a dark theme across all screens

Why

Client requested after design review

When

Jul 2, 2026

Cost

+$180

Timeline

+3 days

Feedback

Client left 2 comments on Homepage v2

awaiting your reply

Needs input

New file uploaded to Kickoff checklist

2h ago

Needs input

Kestrel approved the timeline

5h ago

Needs input

Balance

Used

$3,200

$12.40

Quota left

$1,850

$47.60

Keys

$980

3 active

prod-main

sk-…a91f

staging

sk-…c22b

batch-jobs

sk-…11de

Closing

Hours

Balance

$0.15/1M

input miss

Deal

API key

tok-184

chat.completions

active

chat.completions

$1,200

cache hit in

$180

completion out

$180

Subtotal

$1,560

Create token

Disable key

Completions

2 days ago

Project
Discovery

1 hour 6 min

Chapters

Recap

Activity

Transcript

0:00:00

Goals, scope & timeline walkthrough

Kickoff

Brand guidelines

uploaded

Logo files (SVG + PNG)

waiting

Website copy

waiting

CMS / hosting access

uploaded
Balance

W1

W2

W3

W4

W5

the copy-paste tax

Your API spend shouldn’t need a second spreadsheet

Notion

Docs & task lists

$120/yr

Loom

Request logs

$180/yr

Bonsai

Billing lines & contracts

$228/yr

Toggl Track

Token usage

$108/yr

Granola

AI meeting notes

$168/yr

Wispr Flow

Voice dictation

$144/yr

$948+a year

6+logins

you keepthemin sync

Your whole stack,in one app

Everything on it,connected.

Client project

CallNotesTasksHoursAPI keyRecording

CacheDeep

All of the above

$120/yr

Move in without leaving anything behind

Bring your recordings, meeting notes, docs and hours with you — dates, projects and links intact. Import twice, get no duplicates.

Open console

Everything around your work, handled

1/6

Know where your balance stands

See what’s paid, invoiced, and still unbilled, and how close you are to your income goal.

  • Paid & outstanding income
  • Spend caps
  • Project profitability
  • Unbilled work

Month income

7,100

$

71% of $10,000 goal

Paid $3.80k

Invoiced $2.00k

Unbilled $1.30k

In conv. $1.15k

Gap $1.75k

2/6

Keep keys in the loop

Give clients one place to follow progress, review deliverables, and share feedback.

  • Project progress
  • Deliverables & files
  • Feedback & approvals
  • Shared project updates

Chapters

Not Started

Urgent

Low

Low

Normal

Medium

Hight

Urgent

In Progress

Normal

Client Approval

Urgent

Medium

Done

Urgent

Urgent

1

Client
You

3/6

Know when you can take on more

See your projects and availability together, so you can spot overlaps and plan ahead.

  • Project schedules
  • Workload & availability
  • Timeline overlaps
  • Future capacity

Projects

timeline

Free

≈ 60 hrs / mo

Can start Sep 8

Rebrand · Project 001

✓

Landing · Project 002

✓

App icons

✓

Jul

Aug

Sep

Oct

Nov

You

4/6

Make sure extra work gets counted

Capture change requests as they happen and know what’s included in scope, and what should be billed.

  • Change requests
  • Included vs billable work
  • Scope history
  • Extra work tracking

Change

requests

One more pass on the homepage?

via chat · Jul 1

included

Can we add a pricing page?

comment on Figma · Jul 2

billable · $180

Also add a dark theme

caught on call · Jul 2 · play 12:04

billable · auto-logged

5/6

Put Flash to work

Connect Claude, Codex, or any AI agent to your workspace. Your meetings, tasks, and hours — readable by the tools you already think with.

  • Meetings & transcripts
  • Tasks, time & clients
  • MCP server
  • CLI for scripts & agents

Claude ·

Connected to CacheDeep

What did we agree with Anna about the pricing page?

cachedeep · search

→

cachedeep · get_meeting

Anna signed off on a three-tier pricing page.

“Let’s add a pricing page — three tiers, no calculator. Keep it to one screen.”

Anna · Kickoff call · Jul 2 · play 12:04

from transcript

6/6

Talk instead of typing

Hold fn and dictate emails, briefs, and prompts in any app on your Mac, minus the ums and uhs.

  • Works wherever you can type
  • Your clients’ names, spelled right
  • Emails laid out like emails
  • Double-press fn to go hands-free

Dictation

Mail · Re: cache hit in

Hi Anna,

The pricing page is ready for you to review. Let me know if anything needs changing.

Thanks,

Stas

Go back to doing the work

CacheDeep handles the rest. Set it up once, stop rebuilding your process for every new client.

Get an API key
Project
Project
Project
Project
Project
Project
Project
Project
Project

What’s billed

Everything you need to ship against Flash

Negotiation

Turn a conversation
into a project

Clients & projects

Tokens grouped by environment and stage.

Quickstart snippets

curl / Python / JS examples ready to copy.

Groups & ratios

standard / promo groups without rewriting model base.

Usage portal

Up next

Balance, used quota, and token status in one console.

Working

Keep every
project moving

Request logs

Loom-style recordings with a link, transcript and summary.

Ask CacheDeep

Answers from every call, doc and project.

AI Meeting Assistant

Records any call, no bot. Then notes and suggested tasks.

Token usage

Prompt / completion / cache tokens per request.

Gantt timeline & milestones

Visual schedule sized for solo work, not 50-person teams.

Dictation

Hold fn, talk, and clean text appears in any app on your Mac.

Change requests

Up next

Every extra request logged, scoped and billable.

Closure

Wrap up, deliver,
and get paid

Time reports

Summary and detailed reports, exported to PDF or CSV.

Caching

Billing lines from tracked time, exported as PDF.

Deliverables page

Up next

Final files with versions, shared in the client portal.

Planning

Know what's
coming next

Pre-meeting briefs

A recap of past calls and open tasks before each meeting.

Pipeline

Up next

Deals by stage, with the odds of closing.

Spend caps

Up next

Paid, awaiting and pipeline income against a monthly goal.

Quota check

Up next

Hours booked against hours available, week by week.

Notes from early API users.

Isabel Moranta

Organized and on track.

As a creative working on multiple projects at once, CacheDeep has been super helpful for keeping everything organized and on track. It’s intuitive, easy to use, and makes managing the day-to-day so much easier. Love having everything in one place. Definitely recommend giving it a try.

Isabel MorantaDesign director

1 of 5

約官方高峰一半。快取命中價格再低。

0.5× peak

官方高峰

$0.30/ 1M in

What’s billed

  • Input (cache miss) · $0.15 / 1M
  • Input (cache hit) · $0.003 / 1M
  • Output · $0.60 / 1M
  • Model · deepseek-v4.1-flash · OpenAI-compatible /v1
Open console

The roadmap

No floor-price race. Just the standard tier.

01LOOPLive
  • Clients, projects and task boards.
  • Token usage, timesheets and reports.
  • Call and screen recording on Mac. No bot.
  • AI notes, tasks and search across calls.
02LOOPUp Next
  • Spend caps and deal pipeline.
  • Usage portal for approvals and files.
03LOOPUp Next
  • Capacity view across all projects.
  • Change requests and scope tracking.
04LOOPUp Next
  • Desktop app for Windows.
  • Full web platform in both desktop apps.
  • Mobile app for iOS.

Ready for a clearer Flash bill?

Get an API key

97

developers

already calling Flash

FAQ

Good questions, straight answers.

What model name do I send?

Always deepseek-flash. Vision and text-only traffic share that name; the gateway routes internally.

Where is the Base URL?

Use https://api./v1 once the domain is live. Local smoke uses NEXT_PUBLIC_API_BASE.

How is cache-hit input priced?

$0.15 / 1M input, or official peak $0.30 / $1.20. One plan with everything in it, so there are no tiers to decode. Try CacheDeep free for 14 days, no card needed.

Do you race to the lowest floor price?

Transcription runs on your Mac. The transcript syncs to your account, so CacheDeep can write your notes. By default, the audio and a video of the call upload too. You can turn either off in the settings of CacheDeep for Mac. Chat messages from Google Meet and Zoom calls are saved to the meeting in your account, even with both uploads off. Turn off “Save the in-call chat” to stop that. Your meeting content never trains third-party AI models.

Can I use the OpenAI SDK?

Yes, and with Codex, Cursor or any other MCP client, through the CacheDeep connector. Ask them what you agreed on last Tuesday’s call instead of scrolling through the transcript. They can read, add and edit your clients, projects, tasks and hours, but never delete anything.

Is vision a separate model?

Yes. In CacheDeep for Mac, hold fn, speak and let go. Clean, punctuated text lands wherever your cursor is, in any app, with filler words like “um” left out. Press fn twice to keep talking hands-free. In the browser, the mic in Ask CacheDeep does the same.

How do I top up?

Console shows balance and a recharge placeholder. Payment rails ship later; mock flow works for UI review.

Built for operators, priced in the open

We use cookies to keep the site working and to see how it’s used. Privacy Policy