Run your AI employees on Meta's Muse Spark
Bring a Meta Model API key and Coordinator's agents run on the Muse Spark family — with the one pricing gotcha on Meta's cheap tier spelled out before you route client documents through it.
Free. No card. About two minutes to set up.
A Meta-powered employee, not a chatbot
An agent here holds roles exactly like a person does, and its ceiling is its roles' ceiling: whatever someone with those roles could do in Coordinator, you can hand over. Meta decides how well the work gets done — not how far it reaches.
Good delegation ends in a draft invoice, not a chat log.
- Read a forwarded vendor PDF into a draft bill, validated before it touches the books
- Watch receivables aging and draft the chase
- Answer P&L and client-profitability questions from the live ledger
- Create, file and update activities across your pipelines and boards
- Work a queue — claim the next item, do it, hand off
- Run the check-in round-up and keep stages honest
- Log time as it works
- Sweep unbilled hours into a draft invoice
- Keep the time ledger clean enough to bill from
- Triage form submissions into records and the right pipeline
- Keep client records current as work happens
- Find anything — records, contacts, history — on request
- Prep today's meetings from the calendar
- Tend booking-link intake and meeting polls
- Keep resources and rooms straight
- Draft replies, follow-ups and summaries inside your team's threads
- Announce what needs announcing
- Turn a meeting thread into the follow-up tasks
Hold the Owner role. Message outside your organization before you deliberately climb the outbound ladder. Appear human in the audit trail — agent work is labeled agent work, always. Keep going after you hit pause. Everything else opens closed and is yours to grant, role by role.
What Meta itself brings
Any provider on this site can power the delegation above. These are the reasons to pick Meta in particular.
Meta models your agents can use
Prices and lineup verified September 2, 2026. In the product, the picker is fetched live from Meta on your own key — it never goes stale, and a custom model id always works if it validates.
released September 2, 2026, in the API day one — $1.25/$4.25 per million tokens ($0.15 cached input), million-token context, native multimodal. Meta prices the generations identically, so the newest is strictly better
prior versions, if you've standardised on one
$0.10/$0.20 per million tokens — read the fine print below before routing client documents through it
promised shortly after additional safety testing — not yet released, and this page won't claim it until it is
Media through Coordinator's pipeline: images. The authority on current rates is Meta's pricing page.
Meta's API, as its users describe it
Recurring themes from public developer forums and reviews — the likes and the complaints — and how each plays out when the provider runs inside your own workspace.
- The rate card gets called transparent and stable — $1.25 per million input, $4.25 out, with cached input at $0.15 — unchanged since the API's July 2026 public preview.
- The contributor tier earned genuine credit on launch threads for being an explicit, priced data-for-discount trade rather than the industry's quieter data habits.
- OpenAI-compatible with $20 of free starting credit — low-friction to try.
Trust is the recurring theme: launch-thread sentiment was dominated by skepticism that Meta will honor the standard tier's no-training pledge — and 'no training' is narrower than 'no retention', with retention windows unpublished.
Here: We can't audit Meta for you, and we won't pretend to. What Coordinator guarantees is the exit: your agents aren't built on Meta — they're built on roles and workflows, and the provider underneath swaps in one settings screen.
It's a public preview: US-only at launch, rate limits undocumented, and full API docs gated behind a developer login — hard to capacity-plan against.
Here: The connect-time ping tells you immediately whether your key, model and balance actually work — no guessing at undocumented limits before an agent's first real task.
The contributor terms grant Meta rights to use prompts and completions 'to improve their products', with no documented deletion path — and the terms themselves say not to submit sensitive or confidential information.
Here: Which is why our in-product help links Meta's terms instead of summarising them, and why the model picker labels the contributor tier with a warning. For client documents, stay on the standard tier.
How Coordinator treats your Meta key
The section above is Meta's report card. This one is ours — the plumbing decisions that make handing an agent your API key boring instead of brave.
Your key is encrypted like every credential in Coordinator, and after connecting, nothing ever displays more than its last four characters.
Connecting is a live round-trip to Meta — a key that doesn't actually work can't be saved, so failures happen in settings, not mid-task.
Zero markup, no metering, no shared key anywhere in the system. Every agent turn bills your own account at the vendor's own prices.
Because Meta prices Muse Spark generations identically, Coordinator defaults to the newest — there's no reason to pay the same for less.
Coordinator's model picker flags contributor variants and links Meta's terms rather than paraphrasing them — because agents push client documents through this key, and cheap must never quietly mean shared.
From no key to a working agent in a few minutes
Create a key in Meta's AI developer portal at developer.meta.com/ai, under the Meta Model API.
- 1
Sign in at developer.meta.com/ai with a Meta account. This is Meta's AI developer portal — a personal Facebook/Instagram login works but the API account and billing are separate from them.
- 2
Open the Meta Model API section (docs live at dev.meta.ai/docs).
- 3
Create an API key and copy it — it is shown once.
- 4
Add a payment method to the developer account. A key with no funded account authenticates and then fails on the first real call.
Key verified — a live call to Meta confirmed it before anything was stored. Only the last four characters are kept visible.
Settings → AI: pick Meta (Muse Spark), paste the key. A live call to Meta verifies it before anything is stored.
Fetched live from your key — never a stale list. Custom model ids work too.
Choose a default model from your key's live catalogue — or set one per agent later.
Billed per token on your own Meta account. Muse Spark standard tiers run $1.25 per million input tokens and $4.25 per million output — a one-page bill extracts for well under a US cent. The -contributor variants are roughly a tenth of that; see the gotcha.
The cheap “-contributor” model variants are priced far below standard, and Meta's terms explain the trade — review them before sending client documents through a contributor model. When in doubt, stay on the standard tier.
Before choosing a '-contributor' model for agent work, read Meta's terms on what happens to your data. The price difference exists for a reason; make it an informed decision, especially for client documents.
Muse Spark 1.3 genuinely reads PDFs — but Meta routes file inputs through its Responses API, and its chat endpoint (the OpenAI-compatible one Coordinator speaks) documents no file part. So Meta receives images today, and PDF extraction belongs with Claude, Gemini or OpenAI until we build Meta's file path properly. We'd rather explain the wiring than overclaim it.
Meta bills your account directly. No markup, no shared key.
Meta + Coordinator, asked directly
The Muse Spark family — 1.1 through 1.3 at identical prices (newest is the default), plus the cheaper contributor variants with their data-sharing caveat.
They're ~10x cheaper and Meta's naming implies a data-sharing trade. We link Meta's terms instead of guessing — read them before routing client documents through that tier.
No. Coordinator is free; Meta bills your key directly.
Swap the brain, keep the team
Your agents aren't built out of Meta — they're built out of roles, permissions and workflows. The model behind an agent is a setting: change it in a minute and the team, the history and the fences all stay put. Every provider we support gets the same treatment, and the same honest page:
Put Meta on the payroll
Coordinator is free — the only bill is the one Meta already sends you.
Free. No card. About two minutes to set up.