Which models can I use?

Frontier models from Anthropic, OpenAI, and Google, plus the most powerful open-source models — DeepSeek, Qwen, Kimi (Moonshot), and GLM — along with MindsHub Air, the always-available free-tier baseline. Pick a planning model and a coding model from a dropdown; the Model Router is pre-wired, so there are no provider API keys to set up.

Why run agents on open-source models?

Agents consume far more tokens than a chat session — they plan, call tools, read results, and iterate over many steps. Running that volume on cost-effective open-source models cuts the bill dramatically, and you can still reach for a frontier model on the steps that need it.

If I switch models, do I lose my work?

No. The model is a setting under your agent. Switch it and the same agent keeps running — your tasks, history, and memory carry over. There is nothing to migrate or re-set-up.

No. The Model Router is wired in across commercial and open models, so there are no provider keys to manage. You can connect your own LLM accounts later if you want direct provider control.

What are the latest:* aliases?

Instead of pinning a specific model version, reference an alias like latest:deepseek or latest:gpt. The Model Router resolves it to the current pinned version of that family, so you pick up new releases without changes.

What happens if a model provider has an outage?

The Model Router fails over automatically. If your chosen model is slow, rate-limited, or down, the request retries on a fallback model — often on a different provider — so your task keeps running. latest:mindshub_air is always available as the baseline model, even during an outage.

Model Router

Every model, pre-wired. Swap anytime.

Q: Can I pick a different model per task?

Today the model applies across your work. Choosing a different model — and a different agent — per individual task is coming soon.

Agents burn through tokens to get real work done. Run them on efficient, low-cost open-source models to keep spend under control, then switch to a frontier model the moment a task needs more power — all pre-wired, swap anytime.

Get your API key

MindsHub

live preview

Let's knock things off your list

Agent settings: Default

Q2-Reporting

Build an interactive KPI app the team can explore

Build me an interactive app the team can use to explore our KPIs — MRR, signups, NRR — pulling live from Postgres and Stripe. If something looks off with signups, let us open a Linear ticket and assign it right from the app. Publish it so the team can dig in.

PgPostgreSQLSStripeLLinear

Planned the analysis3 KPIs · live refresh · drill-down by segment
Connected to PostgreSQLvia the credentials vault — no raw key exposed
Queried Stripe for MRR & signups1,204 subscriptions reconciled
Wired up Linear for one-click assignmentissues created straight from flagged rows
Published kpi-app.htmllive, shareable URL

Done — here's your KPI app. MRR is up +9.4% week-over-week to $128.4k. One thing worth a look: signup checkout is erroring for EU accounts — I flagged it below so someone can grab it.

MRR, signups, and NRR update live — nothing to re-run.
Flagged a signup issue for EU checkout — assign it to a teammate right from the app.
Published to a live URL — the team can explore and assign follow-ups without asking you first.

Revenue-Ops

Top customer-feedback themes, ranked by revenue

Combine last month's Gong calls, Zendesk tickets, and HubSpot notes into the top customer-feedback themes — and rank them by revenue impact, not just frequency.

GGongZZendeskHHubSpot

Pulled 42 Gong call transcriptsMay 1 – May 31
Read 318 Zendesk ticketsplus 96 HubSpot deal notes
Clustered into 7 themessemantic grouping + dedupe
Scored each theme by attached ARRcross-referenced HubSpot deals
Wrote the memofeedback-themes.md

Here's the May feedback memo. Ranked by revenue, the #1 theme is onboarding friction — 28 mentions across $214k of ARR, well ahead of where raw frequency would rank it. Pricing confusion and a missing Salesforce sync round out the top three.

Every theme links back to the exact calls, tickets, and quotes.
I drafted a short product follow-up list at the end of the memo.

general

Extract 38 receipts into a Q2 expense workbook

Extract these 38 receipts from the shared Drive folder and build a categorized Q2 expense workbook with monthly totals. Flag anything that looks like a duplicate.

△Google Drive

Read 38 receipts from Google DriveOCR on PDFs and photos
Normalized vendors & currencies3 currencies → USD
Categorized into 12 bucketssoftware, travel, meals, …
Flagged 2 likely duplicatessame vendor, amount, ±1 day
Built expenses.xlsx3 sheets, formulas intact

Built expenses.xlsx — 38 receipts totaling $24,318 across 12 categories. Software is the biggest line at $9,240. I flagged 2 duplicate charges ($318) for your review on the Summary tab.

Each row links to the original receipt in Drive.
Monthly totals and category subtotals are live formulas, not pasted values.

general

Daily 7am briefing across Slack, Gmail, Calendar, Linear

Every weekday at 7am, scan Slack, Gmail, my Calendar, and Linear, and give me a priority briefing before my first meeting — what changed overnight and what needs me.

#SlackMGmail31Google CalendarLLinear

Scanned 6 Slack channelsovernight messages + mentions
Triaged 23 new emailsurgent vs. FYI
Read today's calendar5 meetings, 2 need prep
Checked Linear4 issues moved to your review
Scheduled the recurring runweekdays · 7:00 AM

Scheduled ✓ — I'll run this every weekday at 7:00 AM and drop the brief here. Today's headline: 3 things need you before 10am. Acme is waiting on your reply, the pricing doc needs sign-off, and your 9:30 needs 5 minutes of prep (notes attached).

Everything links to the source — one click to the Slack thread or email.
You can change the cadence or what it scans anytime in Scheduled Tasks.

general

Size the US SMB-analytics market and compare vendors

Size the US SMB analytics market, compare the top vendors by price and ICP, and build me a shareable briefing I can send to the team.

BQBigQuery

Researched 14 market sourcesreports, filings, pricing pages
Modeled TAM / SAM / SOMbottoms-up by segment
Compared 8 vendorsprice, ICP, and gaps
Compiled a sourced appendixevery number cited
Built market-sizing.htmlpublished, shareable

Here's the market-sizing briefing. TAM $18.4B, SAM $6.2B, three-year SOM $430M. I compared eight vendors — the clearest wedge is sub-50-seat teams that enterprise BI tools price out and overserve.

Every figure carries a citation in the sources appendix.
Published to a live URL — share the link, no export needed.

Projects

Group work, set instructions, and keep its data and artifacts together.

general

Default project

3 tasks2 artifacts

Q2-Reporting

Board metrics & dashboards

1 tasks1 artifacts

Revenue-Ops

GTM signal & feedback

1 tasks1 artifacts

Scheduled Tasks

Cowork runs these on the cadence you set, then drops the results in the task.

Schedule anotherWeekly summaries, recurring audits, morning digests…

Live Artifacts

Documents, dashboards, and apps Cowork produced. Publish to share a live URL.

5 artifacts

Connected Apps & Data

Connect a source once. Secrets go into a vault, scoped per connection — the agent never sees a raw key.

Connected

Pg PostgreSQLDatabases Connected

S StripeFinance Connected

M GmailProductivity Connected

# SlackProductivity Connected

31 Google CalendarProductivity Connected

L LinearEngineering Connected

Available — featured

Memory

Rules, lessons, and saved context Cowork can reuse — yours to inspect and edit.

Global 1

Project · general 2

Select a memory file to inspect it.

lessons.md global

— Always reconcile Stripe MRR against the subscriptions table before reporting; they drift by ~2%.

— The boss prefers numbers first, then the narrative. Lead with the headline metric.

— Publish dashboards as live URLs by default — the team shares links, not screenshots.

lessons.md project · general

— Fiscal quarter starts in February, not January. Q2 = May–Jul.

— "North-star" metric is NRR; report it on every dashboard.

rules.md project · general

— Never write to production databases. Read-only credentials only.

— Currency is USD. Convert at the receipt date's rate.

Skill Library

Teach a skill once; Cowork recalls it when the work fits — no relearning.

Settings

Cowork configuration and workspace preferences.

Providers

MindsHub Model Router

Routes via MindsHub with smart model selection. Pre-wired — no per-provider API keys.

•••••••••••••••••• Connected

Agent settings

Planning model

Reasoning, orchestration, and responses.

Model

Reasoning effort

Coding model

Scratchpad code generation.

Model

Harness

Which AI agent powers your tasks.

Estimated performance Updates with the models and reasoning effort

Quality—

Speed—

Cost—

Appearance

Theme

Light or dark — also drives the animated background.

Animated background

Toggle off for a flat surface instead of the moving grid.

created with MINDSHUB

POSTGRES + STRIPE + LINEAR · LIVE

Growth KPI App

Live · updating now

MRR

$128.4k

▲ 9.4% WoW

New signups (24h)

412

▲ 6.0%

Churn

2.1%

▲ 0.3 pt

Net revenue retention

112%

— flat

Signups per hour, last 48h — click a red dot to explore

65/hr

Requests per minute

842/min

Recent signups

RESEARCH · 14 SOURCES

US SMB Analytics — Market Sizing

Published · shareable

TAM$18.4BAll US SMB analytics spend

SAM$6.2BSelf-serve, < 200 seats

SOM$430MReachable in 3 years

Vendor comparison

Vendor	Entry price	ICP	Gap
Enterprise BI A	$70k/yr	2,000+ seats	Overserves SMB
Dashboard SaaS B	$24/mo	10–50 seats	No warehouse
Notebook tool C	$0 OSS	Analysts	Not self-serve
The wedge →	—	< 50 seats	Underserved

Sources: Gartner 2025, vendor 10-Ks, pricing pages, Census CBP. Every figure cited in the appendix.

Customer Feedback Themes — May

Prepared by MindsHub · sources: 42 Gong calls, 318 Zendesk tickets, 96 HubSpot notes

Ranked by attached ARR, not raw frequency. Each theme links to its source calls, tickets, and quotes.

1 · Onboarding friction$214k ARR · 28 mentions
2 · Pricing confusion$156k ARR · 19 mentions
3 · Missing Salesforce sync$98k ARR · 12 mentions
4 · Slow CSV export$41k ARR · 22 mentions

"We almost didn't make it past week one — setup took three calls with support." — VP Ops, $48k ARR account (Gong, May 14)

Recommended follow-ups

Ship a guided first-run; target time-to-first-value < 10 min.
Rework the pricing page around seats, not credits.
Prioritize the Salesforce connector — 5 deals waiting.

Daily Briefing — Friday

Generated 7:00 AM · Slack · Gmail · Calendar · Linear

Reply to Acme — they're waiting on the renewal terms.
#sales · 7:42 PM
Sign off on the pricing doc before the 11am review.
Gmail · Dana
Your 9:30 with Initech needs 5 min of prep.
Calendar

Meeting prep · 9:30 Initech

Renewal call. They asked about SSO last time — it shipped Tuesday. Up-sell angle: 15 seats added in May.

Open loops

4 Linear issues moved to your review overnight.
2 Slack threads in #product mention you.

Every line links back to the source — one click to the thread, email, or issue.

expenses.xlsx ⚑ 2 duplicates flagged

SummaryReceiptsCategories

	A · Date	B · Vendor	C · Category	D · Amount	E · Note
1	05-03	Linear	Software	$1,200.00
2	05-07	AWS	Infrastructure	$3,480.12
3	05-09	Delta	Travel	$612.40
4	05-12	Figma	Software	$144.00	⚑ possible duplicate of row 9
5	05-15	Notion	Software	$96.00
6	05-18	Uber	Travel	$58.20
7	05-21	Sweetgreen	Meals	$84.10
…		TOTAL		$24,318.00	=SUM(D2:D39)

Interactive preview — click around, open an artifact, or type a task.

Try it — pick a planning model and a coding model, tune the reasoning effort per model, and watch the estimated performance update live. The Model Router is pre-wired, so there's nothing else to set up.

Why it matters

The right model for each job — without the busywork.

Model choice is where agents quietly win or lose on cost and quality. MindsHub makes it a dropdown, not a project.

Cost

Control spend with open-source models

Agents burn through far more tokens than a chat session — they plan, call tools, read what comes back, and iterate over many steps. Run that volume on efficient, low-cost open-source models like DeepSeek, Qwen, and Kimi, and the bill drops sharply versus frontier pricing.

Power on demand

Reach for a frontier model when it matters

Some steps deserve the best reasoning available. Switch to a frontier model for the hard part and back again when you're done — the agent stays the same, every task still runs, and your history and memory carry over. Nothing to migrate.

Control

One place to manage it — no API keys

The Model Router is pre-wired across providers, so there's nothing to set up. Pick a planning model and a coding model from a dropdown, set the reasoning effort, and you're done — no juggling provider accounts or per-provider keys.

Per task Coming soon

Swap models on the task level

Today the model applies across your work. Choosing a different model — and a different agent — for an individual task is on the way, so you can match each job to the right engine.

The lineup

From frontier to open-source — one router.

Mix commercial and open models freely. The Model Router sends each step to the model you choose and keeps versions current with latest:* aliases.

Anthropic Frontier
OpenAI Frontier
Google Frontier
DeepSeek Open-source
Qwen Open-source
Kimi (Moonshot) Open-source
Grok (xAI) Frontier
Muse Spark (Meta) Frontier

Every alias

Frontier and open-source, ready to call.

Reference any model by its latest:* alias — it stays current as providers ship new versions.

curl https://api.mindshub.ai/v1/chat/completions \
  -H "Authorization: Bearer $MINDSHUB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "latest:sonnet",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Open sourceOpen weights

Alias	Resolves to	Backup
`latest:kimi`	`kimi-k2p6`	`latest:gemini-flash`
`latest:deepseek`	`deepseek-v4-pro`	`latest:gemini-flash`
`latest:qwen`	`qwen3p6-plus`	`latest:gemini-flash`
`latest:glm`	`glm-5p2`	`latest:gemini-flash`
`latest:mindshub_air`	`kimi-k2p6`The free-tier baseline model	`latest:gemini-flash`

AnthropicFrontier

Alias	Resolves to	Backup
`latest:fable`	`claude-fable-5`	`latest:gemini-flash`
`latest:sonnet`	`claude-sonnet-5`	`latest:gpt`
`latest:opus`	`claude-opus-4-8`	`latest:gpt-high`
`latest:haiku`	`claude-haiku-4-5-20251001`	`latest:gemini-flash`

OpenAIFrontier

Alias	Resolves to	Backup
`latest:gptlatest:gpt-sollatest:gpt-lowlatest:gpt-medium`	`gpt-5.6-sol`Flagship. Low or medium reasoning effort	`latest:sonnet`
`latest:gpt-high`	`gpt-5.6-sol`Flagship. High reasoning effort	`latest:opus`
`latest:gpt-terra`	`gpt-5.6-terra`Balanced mid-tier	`latest:sonnet`
`latest:gpt-luna`	`gpt-5.6-luna`Fast and economical	`latest:gemini-flash`
`latest:gpt-codex`	`gpt-5.3-codex`	`latest:sonnet`
`latest:gpt-mini`	`gpt-5.4-mini`	`latest:gemini-flash`
`latest:gpt-nano`	`gpt-5.4-nano`	`latest:gemini-flash`

GoogleFrontier

Alias	Resolves to	Backup
`latest:gemini`	`gemini-3.1-pro-preview`	`latest:sonnet`
`latest:gemini-flash`	`gemini-3.5-flash`	`latest:gpt-mini`

Reliability

If a model goes down, MindsHub doesn't.

Every request runs through automatic failover. If a provider is slow, rate-limited, or down, the Model Router retries on another model in the background — your task keeps moving without you noticing.

Automatic, not manual

No error message, no re-run. If the model you picked can't respond, the Model Router retries the request on a fallback model automatically.

Cross-provider by design

Fallback chains span providers — an Anthropic slowdown can fail over to OpenAI or Google, not just another model from the same vendor.

MindsHub Air, the always-on baseline

latest:mindshub_air ships on every plan and stays available even when other models are under strain — a guaranteed model that is always ready to pick up the work.

FAQ

Questions about models.

Which models can I use?: Frontier models from Anthropic, OpenAI, and Google, plus the most powerful open-source models — DeepSeek, Qwen, Kimi (Moonshot), and GLM — along with MindsHub Air, the always-available free-tier baseline. Pick a planning model and a coding model from a dropdown; the Model Router is pre-wired, so there are no provider API keys to set up.
Why run agents on open-source models?: Agents consume far more tokens than a chat session — they plan, call tools, read results, and iterate over many steps. Running that volume on cost-effective open-source models cuts the bill dramatically, and you can still reach for a frontier model on the steps that need it.
If I switch models, do I lose my work?: No. The model is a setting under your agent. Switch it and the same agent keeps running — your tasks, history, and memory carry over. There is nothing to migrate or re-set-up.
Do I need API keys?: No. The Model Router is wired in across commercial and open models, so there are no provider keys to manage. You can connect your own LLM accounts later if you want direct provider control.
Can I pick a different model per task?: Today the model applies across your work. Choosing a different model — and a different agent — per individual task is coming soon.
What are the latest:* aliases?: Instead of pinning a specific model version, reference an alias like latest:deepseek or latest:gpt. The Model Router resolves it to the current pinned version of that family, so you pick up new releases without changes.
What happens if a model provider has an outage?: The Model Router fails over automatically. If your chosen model is slow, rate-limited, or down, the request retries on a fallback model — often on a different provider — so your task keeps running. latest:mindshub_air is always available as the baseline model, even during an outage.

Every model, pre-wired. Swap anytime.

Let's knock things off your list

Projects

general

Q2-Reporting

Revenue-Ops

Scheduled Tasks

Live Artifacts

Connected Apps & Data

Memory

lessons.md global

lessons.md project · general

rules.md project · general

Skill Library

Settings

The right model for each job — without the busywork.

Control spend with open-source models

Reach for a frontier model when it matters

One place to manage it — no API keys

Swap models on the task level

From frontier to open-source — one router.

Frontier and open-source, ready to call.

Open sourceOpen weights

AnthropicFrontier

OpenAIFrontier

GoogleFrontier

If a model goes down, MindsHub doesn't.

Automatic, not manual

Cross-provider by design

MindsHub Air, the always-on baseline

Questions about models.

Pick your models. Change your mind anytime.