Skip to main content

Describe your business in plain English. A short conversation surfaces the best-match template mid-chat and assembles a coordinator and specialists — tailored to your industry — that hand off to each other, run on a schedule, and pause for your approval before anything risky.

Start freeSign in
  • Row-Level Security (RLS) tenant isolation
  • Bring your own model key (BYOK)
  • Receipt-backed actions
  • $0 to start, no credit card

Powered by Claude, GPT, Gemini, Grok, Kimi, and DeepSeek. Pick the right brain per task — or bring your own.

Book a 15-min demo →

What sets it apart

Six things nothing else does together.

01

Receipts, not vibes.

When an agent says it sent the email, the platform checked. Risky actions — outbound sends, published content, filed approvals — produce typed runtime receipts, and a run can’t settle as done until every effect it claimed is confirmed against what actually happened. If the send failed, the run doesn’t get to claim it succeeded.

02

A team that spawns its own sub-agents.

A coordinator routes work to specialists who hand off to each other — and mid-run any agent can spawn sub-agents that work in parallel, then joins their results before it continues. Risky moves still wait for your approval; the Atlas run forest shows every parent and its sub-runs.

03

Auto-tunes the model per task.

You don’t pick a model — set the org default to Auto and the router picks the best one per task within your exclusions, escalating off any unavailable provider automatically. Exclude a whole provider or a single model, pin or override per agent, or bring your own key.

04

Repairs its own platform.

A daily proactive service scans every org for stuck crons, broken MCP servers, expired OAuth, and stalled approvals. It auto-applies the mechanical fixes and flags the rest as operator actions — the platform heals itself before the failure reaches you.

05

Has a built-in support agent.

A floating chat widget on every page reads your real org/team/agent config — not a generic FAQ — and proposes a specific fix you approve in one click. It surfaces open findings from the daily review right on the bubble.

06

Adversarial verification for high-stakes work.

When an agent proposes a legal contract, security audit, or investor narrative, the platform opens an approval card with verification in progress. The card stays blocked until the evaluator gate, red-team/author defense, and mechanical checks finish.

  • Conversational, plain-English setup
  • Free to start, no credit card
  • Bring your own API key
  • Row-Level Security isolation
How you start

Three steps from zero to a running team.

  1. Describe your business

    Plain English. A couple of sentences of context.

  2. Pick or accept a template

    34 public ready-made teams. The wizard suggests the right one.

  3. Approve risky moves

    Agents pause before sending money, posting publicly, or anything that needs your eyes.

What’s new

Shipped recently.

Deploy-proof runtime

In-flight agent work survives platform deploys and restarts. Durable job leases track every run, interrupted coordination work lands in a recovery queue, and each recovered step resumes exactly once — nothing lost, nothing run twice.

Exactly-once inbound handling

Every customer message on every channel is claimed by a single owner before an agent handles it, so a busy inbox doesn't drop a message or answer it twice — even mid-deploy.

The Capability Console is live

Integration status is earned from production evidence, not a connect button. The console shows what's wired, what's proven by a real run, and what needs you — in one place.

Self-healing that escalates to you

The platform still auto-repairs mechanical failures. Issues that persist now escalate into a single deduped FYI in your approvals queue — not a buried log line — and recurrences bump a counter instead of spamming duplicates.

Memory that consolidates

Agents reflect on finished work. When the same lesson recurs across agents, it surfaces as a promotion candidate you can approve into a standing directive the whole team follows.

Tier-aware billing margins

Pay-as-you-go bills provider cost + 10%; Pro overage bills at model cost with zero markup. Live budget banners track your month-to-date spend against your cap on the same basis you're billed.

Plug in the knowledge you already have

Your SOPs already live somewhere. So do ours.

Connect Google Drive, OneDrive, Dropbox, or Box once and your agents read everything — with citations on every answer. Per-folder scope; agent-level isolation.

Google Drive

Sync any folder. Files indexed with semantic search. Citations every time.

OneDrive

Sync any folder. Files indexed with semantic search. Citations every time.

Dropbox

Sync any folder. Files indexed with semantic search. Citations every time.

Box

Sync any folder. Files indexed with semantic search. Citations every time.

Pick your sport

34 public pre-configured teams spanning industry verticals plus cross-functional roles.

Industry verticals like Dental Practice, Property Management, and Restaurant Operations plus cross-functional teams like sales, support, finance, growth, and people ops. Every template ships with a coordinator, a hand-tuned roster of specialists, default capabilities, channel presets, and seed knowledge. Start from any one — rename, add, or drop roles freely. The conversational wizard surfaces the best match mid-conversation (“this looks like the Dental Practice template — want to start from there?”) so you don’t have to scroll a list up front.

Property Management

Rent collection, work orders, lease renewals.

SaaS / B2B Startup

Pipeline, customer success, tech support, ops.

Accounting Firm

Engagement intake, document collection, billing.

Dental Practice

Appointment scheduling, recall outreach, billing.

E-commerce / D2C

Orders, returns, inventory, merchandising.

Marketing Agency

Pitch, project mgmt, client billing for agencies.

Real Estate Brokerage

Lead routing, listing prep, transaction follow-through.

Restaurant Operations

Reservations, supplier comms, staff scheduling.

Business Chief of Staff

Inbox, calendar, follow-through, and stakeholder cadence for an owner or exec.

+25 more industries
Coaches / Solopreneurs

Sales, ops, and content for the creator economy.

Auto Repair Shop

Service scheduling, estimates, customer follow-up.

Event Production

Vendor coordination, run-of-show, post-event recap.

Freight & Logistics

Carrier dispatch, exception handling, customer comms.

Gym & Fitness Studio

Member onboarding, retention, churn outreach.

Home Services Contractor

Dispatch, quoting, customer texts.

HR Consultancy

Client intake, candidate sourcing, compliance support.

Manufacturing Job Shop

Quote requests, order management, supplier comms.

Medical Practice

Patient intake, scheduling, follow-ups.

Nonprofit Fundraising

Donor stewardship, grant tracking, volunteer coordination.

Renewable Energy Services

Service intake, audits, rebate workflows, and crew coordination for renewable-energy operators.

Solar Energy

Site qualification, PV production estimates, tariffs, proposals, and installation handoffs.

Salon & Spa

Booking, recall outreach, gift card sales.

Small Law Firm

Intake, client comms, doc collection — non-bar-restricted tasks.

Finance & Bookkeeping

Monthly close, AP/AR, FP&A, and payroll — every disbursement gated on approval.

Software Development

Eng lead, architect, backend, frontend, SRE — approval-gated on every merge + deploy.

Hardware Design

PCB, FPGA, mechanical CAD, BOM/supply-chain across the EVT/DVT/PVT phase gates.

General Operations

Calendar, inbox, vendor coordination, and admin throughput for any 5-50 person team.

Growth Marketing

In-house growth team: content/SEO, paid, lifecycle, analytics + experiments.

Customer Support

Tier-1, Tier-2, KB, self-service — billing changes + customer sends gated on approval.

Research & Analysis

Market sizing, competitor profiles, literature reviews — cite-every-claim discipline.

People & HR

Onboarding, benefits Q&A, comms, recruiting — hard gates on comp + termination.

Personal Assistant Team

Calendar, inbox, travel, and follow-through for an individual principal.

Education & Tutoring

Lesson planning, learner follow-up, parent updates, and tutoring operations.

Trading Desk

Multi-asset research desk — equities, fixed-income, commodities/futures, options, prediction-market events. Simulated paper trades and operator-approved order proposals; never trades real money on its own.

Don’t see your industry? Start from a blank template and we’ll add yours to the queue — every unmatched request shows up in our roadmap, ranked by how often it’s asked.

Instant integrations

50+ MCP plugins. 91 skill packs.

Stripe billing, Notion docs, HubSpot CRM, Linear issues, DocuSign signing, Sentry observability, Atlassian (Jira / Confluence), Shopify, QuickBooks, ServiceTitan, Clio, Mindbody, AppFolio, Gusto, Toast, Salesforce, Clay, Canva, Figma, Asana, Monday.com, Box, WordPress, Ghost, Buffer, X, LinkedIn, Meta, Snapchat, Enphase, SolarEdge, plus vertical-specific MCP wraps and a curated skill-pack library. MCP plugins give your agents tools (what they can do); skill packs give them house style and task plans (how to do it your way). Credential-gated plugins require a connected org credential before activation; wire tools at Settings → Tools and skill packs at Settings → Agent Context.

Browse the catalog
How it works — the playbook

Snap. Block. Tackle. Ship.

One line in; a coordinated drive out. Here’s the playbook your team runs every time.

01 — Snap

Agent takes the snap.

“Close the books for Q3.” “Draft the campaign brief.” “Onboard the new vendor.” One line. The huddle takes it from there.

Context window
Reads everything, holds nothing back

The doc you forgot to share, the spreadsheet from finance, the email thread from last March, the dashboard nobody opens.

Audible
Catches what you don’t say

Picks up on house style — how your org writes, the formats you ship in, the bar a deliverable has to clear before it leaves your desk.

02 — Block

The line holds.

Before the play even starts, two agents are already clearing what would stall it — approvals, missing inputs, conflicting data.

Left guard
Unblocks the inputs

Chases the missing approval, reconciles two versions of the same number, surfaces the policy nobody flagged. The drive doesn’t stall.

Right guard
Risk intercepts

Spots the deadline shift, the regulation change, the stakeholder who needs a heads-up. Drafts the message before you’d think to send it.

03 — Tackle

Hand it off.

The repetitive work — the parts you used to lose half a day to — gets distributed to halfbacks running parallel routes.

Halfback A
Drafts and synthesis

Pulls the brief, the report, the deck, the proposal. Cites where every number came from. Returns it ready for your edit, not a rewrite.

Halfback B
Routing and follow-through

Files the ticket, schedules the meeting, sends the recap, updates the tracker. The trail of small tasks that always slip through.

04 — Ship

Cross the goal line.

The deliverable lands with a clean summary, the open questions called out, and the riskiest pieces already flagged for your review.

Final review
Decisions, not edits

You read a summary, sign off on the calls that needed a human, and approve. The work was already done in the way you’d have done it.

Touchdown
Done before lunch

You started the day with one line. You end it with the report sent, the campaign live, the close booked, the vendor onboarded.

Everything in one huddle

The full playbook.

Five-minute setup, production-grade defaults, customize anywhere you need to. Every team you generate ships wired with these out of the box.

Inside the huddle

Coordination by default.

Agents pass work to each other instead of trying to be heroes. Risky moves wait for your nod. Schedules fire whether you’re online or not.

Handoffs
Coordinated, not chaotic

Inside a team, anyone hands to anyone. Across teams, traffic routes through the coordinators by default. Direct routes are an opt-in for power users.

Approvals
Pause before risky moves

Payments, voice calls, edits to org-wide content, meetings booked with outside guests — agents open an approval card and wait. You approve, reject, snooze or schedule it from one shared approvals queue; the run picks up where it paused.

Scheduler
Proactive, not reactive

Each agent runs on a cron line — Mondays at 9 AM, every morning, end of quarter. Chase invoices, send the digest, open the follow-up tickets without being asked.

Workflows
Multi-stage runs with checkpoints

Out of the box: Adversarial Review (red-team challenge → author defense → operator decision), Roundtable (multi-agent consensus), Counsel (legal lead → finance + security checks → memo), Full Bench (every specialist weighs in). Start any from the Start a workflow button on your Atlas dashboard.

What they know

Grounded in your material.

Your SOPs, your contracts, your style guide, your last six months of email — chunked, embedded, and recalled with citations.

Knowledge cascade
Org → team → agent

Brand guidelines at the org level, playbooks at the team level, the SDR’s cold-email kit at the agent level. pgvector-backed semantic search; every answer cites where the number came from.

Bulk upload
50 files in, review-ready together

Drop up to 50 PDFs, Markdown, or text files at once. The platform reads each one, suggests a title, description, and tags — you review and bulk-approve. No document-by-document grind on day one.

Agent-authored content
They write back

Give an agent the publish tool and it can draft reports, save versions with change summaries, and submit for review. Auto-publish at agent scope; team and org writes always wait for human sign-off.

Edit permissions
Three permission modes

Per-content ACLs, split publish/edit tools, or org-wide free-for-all. Schema carries fields for all three — flip a single column to switch modes; no deploy needed.

Cross-agent precedents
Patterns that spread across the team

When one agent finds a pattern that works — a high-converting outreach hook, a cleaner reconciliation method — it promotes that pattern to a shared precedent board. Every other agent picks it up next run.

Production-ready out of the box

Built for real customers.

The boring stuff handled before you sign up. Multi-tenant isolation enforced at the database. Signed provider webhooks rejected at the door before anything runs. Bring your own provider key if your DPA needs it.

How it’s wired

Isolation by default. BYOK on demand.

Defaults that survive a security review. Designed for the buyer who has to answer to their compliance team before signing.

Tenant isolation
Enforced at the database

Every per-org table has Row-Level Security with deny-all defaults. Application queries are scoped by org id at the query layer. The two checks compound: even an app-layer bug can’t leak rows across tenants.

Bring your own key
Your provider, your billing

Set your own API key per org for any of the six BYOK-eligible providers: Anthropic, OpenAI, Google, xAI, Moonshot AI, or OpenRouter. Calls dispatch through your account, billed by the provider directly. Recommended when your compliance team has a Data Processing Agreement (DPA) with the model provider.

Webhook verification
Fail-closed at every webhook that drives an effect

Stripe, Slack, Twilio, GitHub, AWS SNS, Google Pub/Sub, Box, Dropbox, Alpaca, and Plaidsign their callbacks, and we verify that signature with a timing-safe compare before anything runs — a bad signature is rejected, not logged and processed. Where the protocol carries a timestamp (Slack, Stripe, Box) we also enforce a replay window, and Google Pub/Sub’s token is pinned to our audience. Microsoft Graph, OneDrive, and Telegram publish no signature scheme at all, so those callbacks are bound to a server-generated secret no attacker can forge; Google Drive push notifications carry neither, so we treat them purely as a nudge to re-fetch with our own credentials and trust nothing in the payload. The generic connector dispatcher verifies any provider with a registered scheme; the remaining connectors are read-only, and their inbound webhooks are acknowledged no-ops that mutate no state.

Proven, not advertised
A capability is “working” only once a real run confirms it

Every integration earns its status from production evidence, not a connect button. A capability is marked working only after a real task confirms it ran; one that was working and then broke is surfaced loudly so you find out before your customer does. The Capability Consolemakes exactly what’s wired, what’s proven, and what needs you legible, so there’s no mystery about whether the thing you connected actually works.

Internal security reviews
Findings triaged and closed

Recurring internal security review rounds, including a full vibe-coded-app vulnerability audit exercise, were run against the platform — every HIGH and MEDIUM finding triaged and closed. These were internal exercises, not external penetration tests or independent audits. Live platform health is public at /status.

Pricing

Four ways to run it. No surprises.

Get started free, no credit card. Pool runtime, Dashboard chat, Telegram, plugins, and the full Atlas dashboard work on $0/month. Add a card and pay only for what you use, or upgrade to Pro for a flat rate when your volume calls for it. Need everything inside your own perimeter? Run the Local Platform on your own infra.

Free tier

$0/mo

Shared computing resources · Dashboard + Telegram channels · zero card up front

  • Shared computing resources — your data stays fully separated from every other organization, same as Pro
  • Dashboard chat + Telegram channels
  • All 34 public templates (industry verticals + cross-functional teams)
  • Knowledge base + content authoring with three permission modes
  • MCP server plugins + skill packs (curated catalog + BYO)
  • Atlas dashboard (Triage / Operate / Inspect) with one cross-team approvals queue
  • Org rename, members, activity log, billing portal
Start free
Pay-as-you-go

Pay-per-use

All Pro features · billed monthly for actual LLM compute · add a card when you're ready

  • All Pro channels + features unlocked
  • Auto-routes across 15+ model catalog entries from 5 providers — Anthropic, OpenAI, Google, xAI, Moonshot AI — pin per agent or let the router pick
  • Every model metered at provider cost + 10% margin — e.g. Claude Sonnet 5 $3.30/$16.50, Haiku 4.5 $1.10/$5.50, Gemini 3.1 Flash-Lite $0.28/$1.65 (per MTok in/out)
  • The in-app picker shows the live per-model rate and projected $/mo for your usage before you commit
  • Bring your own API key (BYOK) — no markup, data goes direct to the provider
  • $100/mo default cap (configurable, including unlimited)
  • All paid-tier channels for every team
  • Email, Slack, SMS, WhatsApp, Voice, and webhook channels
  • Add-on marketplace + SSO
Start free, add card later
Pro tier

$49/mo

Flat monthly fee · every channel · add-ons · SSO

  • Everything in Free, plus:
  • All paid-tier channels for every team — owned-domain email, Slack, SMS, WhatsApp, Voice, and webhooks under one plan
  • Model picker — 15+ models across 6 providers, pinned per agent (or auto-routed), or bring your own provider key
  • $40/mo of model usage included — above the budget, marginal usage bills at model cost — no margin — with an in-app banner at 80%
  • First 3 teams included; $10/mo per additional team
  • Email — your own Gmail or M365 inbox via OAuth
  • Slack workspace integration
  • Twilio SMS, WhatsApp, and Voice channels
  • Generic webhooks for any custom HTTP endpoint
  • Add-on marketplace: Hunter.io, SerpAPI, Reducto, Linear (per-unit billing with monthly caps)
  • SSO — SAML 2.0 for your org
  • Priority support — direct line to the team at info@groundpound.ai
Start with Pro
Early access

Local Platform— run the full Pro feature set inside your own cloud, VPC, or air-gapped network. $29/mo for rolling updates or $399 once to own a pinned release. BYOK only; your data never leaves your perimeter. Early access for Pro customers. See the Local Platform →

Add-ons (Pro tier, per-unit billing)
Hunter.ioSerpAPIReducto PDFLinear+ more coming

Each add-on has a per-unit price and a monthly cap you set. Cap exceeded? The agent gets rate-limited and we ping you to lift the cap or pause the add-on. Every charge is logged and reconciled to a Stripe metered line item.

See full plan comparison
FeatureFreePAYGProLocalself-hosted
Dashboard chat + Telegram✓✓✓✓
All 34 public templates✓✓✓✓
Knowledge base + content authoring✓✓✓✓
MCP servers + skill pack plugins✓✓✓✓
Atlas dashboard + approvals queue✓✓✓✓
External Knowledge connectors (Drive / OneDrive / Dropbox / Box)✓✓✓✓
ComputeShared poolShared poolShared poolYour infra
Email channel (own mailbox via OAuth)—✓✓✓
Slack / SMS / WhatsApp / Voice channels—✓✓✓
Add-on marketplace (Hunter / SerpAPI / Reducto / Linear)—✓✓✓
SSO (SAML 2.0)—✓✓✓
5-provider auto-routingAuto, $0 budgetMeteredFlat + budgetBYOK only
Bring-your-own-key (BYOK)✓✓✓✓
Runs entirely in your own perimeter———✓
See per-model rates
Model$/MTok input$/MTok output
Claude Fable 5$11.00$55.00
Claude Opus 5$5.50$27.50
Claude Sonnet 5$3.30$16.50
Claude Haiku 4.5$1.10$5.50
GPT-5.6 Sol$5.50$33.00
GPT-5.6 Terra$2.20$13.20
GPT-5.6 Luna$0.22$1.32
Gemini 3.6 Flash$1.65$8.25
Gemini 3.1 Flash-Lite$0.28$1.65
Grok 4.6$2.20$6.60
Grok 4.20 Reasoning$1.38$2.75
Grok 4.20 Non-Reasoning$1.38$2.75
Kimi K3$3.30$16.50
Kimi K2.6$1.04$4.40
DeepSeek V4 Pro (OpenRouter)$0.48$0.96

All prices are provider cost + 10% margin; Pro overage bills at model cost (no markup). With BYOK, calls dispatch through your account with zero markup. The in-app picker shows the live rate and projected monthly cost before you commit.

Go deeper

Your model. Your stack.

The model question and the integration question, answered. Pick the brain per agent; plug in the rest of your tools without a code change.

Your model, your call

Five providers, one auto-router — picks the right brain per task.

By default you don’t pick a model: set the org default to Auto and the router picks the best model per task within your exclusions, escalating off any unavailable provider automatically so runs still complete. Pin a model or override per agent when you want to, or bring your own key and skip the markup entirely. 15+ production catalog entries across 5 providers, metered at cost + 10%.

15+ catalog entries, one picker
From $0.22/MTok light models up to $55/MTok for flagship output

Auto-routes across 15+ model catalog entries from 5providers — Anthropic, OpenAI, Google, xAI, Moonshot AI. Set the org default to Auto; the router picks the right model per task and falls over to another provider automatically if one is unavailable — or pin a model per agent yourself. Add your own OpenRouter key and its catalog is pinnable too. Every model and its live rate is in the table below.

Run the same team for $1/mo
Gemini Flash-Lite vs. Sonnet: ~10× cheaper

Gemini 3.1 Flash-Lite runs at $0.28/$1.65 per MTok in/out. Claude Sonnet 5 runs at $3.30/$16.50. Same agents, same playbook — pick Flash when throughput matters; pick Sonnet when the output has to be flawless.

Auto-tune by default, pin when you want
The router picks the best model per task

Set the org default to Auto and each task is routed to the best model within your exclusions — telemetry tracks decisions, success rate, and median cost/latency per task signature. Prefer a model until it’s proven, pin one outright, or override per agent — your coordinator on Sonnet, your email halfback on Flash, in the same team.

Bring your own key
Full compliance, zero markup

Have your own DPA with Anthropic, OpenAI, Google, xAI, Moonshot AI, or OpenRouter? Drop your API key at the org level and agents call the provider directly. No metering margin, no data routing through GroundPound infra.

How they extend

Plug in the rest of your stack.

Two extension surfaces and eight channels. Attach at any scope; inheritance stacks. No code change to wire a new external service.

MCP servers
New tools without new code

Plug a Model Context Protocol (MCP) server in (GitHub, Linear, Notion, custom) and your agents gain its tools. Names are namespaced by server slug so two servers’ create_issue tools never collide. Curated catalog plus BYO URL.

Skill packs
House rules in markdown

Best practices, voice and tone, escalation rules, how your company writes emails. Markdown stitched into the system prompt in scope order — org first, agent closest to the user message.

Channels
Meet customers where they are

Dashboard chat and Telegram are free out of the box. Pro unlocks the rest of the lineup: your own Gmail / M365 inbox via OAuth, Slack, Twilio SMS, WhatsApp, Voice, and generic webhooks. A managed mailbox on a GroundPound subdomain (yourorg.groundpound.ai), provisioned via AWS SES, is coming soon. Per-agent overrides — your SDR texts and emails; support takes Slack and email.

Three-scope nesting
Inherit and override

Same rule for tools, skill packs, channels, and content: org defaults, team specialization, agent-level fine-tune. Specific scopes win when they need to; broad scopes apply automatically when they don’t.

Early access For Pro-tier customers

Local Platform — run GroundPound on your own infra.

Run your whole team inside your own cloud, VPC, or air-gapped network — every inbox, document, and embedding stays within your perimeter, dispatched through your own model keys (BYOK only). You bring the keys and the infrastructure; you unlock the full Pro feature set. $29/mo keeps you on rolling updates, or $399 once lets you own and pin a release. Available in early access to Pro customers, who turn on self-host from in-app billing. New to GroundPound? Start with Pro to unlock it →

Local Platform · early access

Run it where your data lives.

Export a portable bundle of your agents, knowledge base, and workflows from Settings → Export, then run the realstack on your own infrastructure with one command — a guided setup wizard brings up the runtime, the dashboard, the database, and a local object store, and runs a parity health check. Pick the billing that fits: a rolling $29/mo subscription or a $399 one-time perpetual license to own a pinned release. Early access for Pro customers.

Local Platform early access
Your cloud, your network, your perimeter

The Local Platform runs the full agent runtime inside your own cloud, VPC, or air-gapped network, with your own model endpoints — so inboxes, documents, and embeddings never leave your perimeter. It unlocks the full Pro feature set, self-hosted and BYOK-only: you bring the model keys and the infrastructure. Available in early access to Pro customers; migrate a managed workspace with the export bundle (Settings → Export).

Two ways to buy
$29/mo rolling, or $399 to own it

Monthly — $29/mo: a rolling subscription that stays current with every release. Perpetual — $399 once: a one-time license to own and pin a release on your own schedule, no recurring fee. Both run the same self-hosted stack with your own model keys. Pro customers turn it on from in-app billing.

Available to Pro customers
Export bundle — the real stack, one command

From Settings → Export you can generate a portable bundle of your agents, knowledge base, and workflows that runs the real stack on infrastructure you control: a guided wizard brings up the runtime, the dashboard, the database, and a local object store, then runs a parity health check. Bring your own keys for the supported model providers, or connect a compatible local model endpoint.

Where they run

Atlas overhead. One pooled runtime.

One operator command center, organized by what you came to do. Every team runs on the shared managed pool runtime, with multi-tenant data isolation enforced in the runtime itself and by continuous CI tests — never a one-row leak between orgs.

Atlas dashboard
Triage, Operate, Inspect

The visit’s purpose picks the layout: Triage for what needs you now, Operate to run and schedule, Inspect to drill into health and cost — plus a narrated Activity timeline and one shared Approvals queue across every team.

Multi-tenancy
Three layers, not one

Database column scoping enforced at every query layer; a signed JWT verified on every Foundry request before any database or model call; and a per-row assertion inside the runtime that the team on the row matches the team on the token, with knowledge-base embeddings scoped to their owner. Multi-tenant data isolation tests run in CI and block deploy.

Free defaults
$0/month to actually run

Pool runtime, Dashboard chat, Telegram, plugins, the full Atlas dashboard, knowledge base + content authoring — all free. Pro or PAYG unlocks the rest of the channels, the add-on marketplace, SSO, and paid model-routing features.

Live cost preview
See the bill before you hit “Generate”

The wizard’s right rail shows a running monthly cost estimate as you add and remove agents. Choose a cheaper model and watch the number drop in real time. Crosses $39.20/mo? The panel surfaces the $49/mo Pro comparison before launch.

IT Support Agent
Diagnoses, recommends, fixes — with your approval

A floating chat widget on every page that reads your actual org/team/agent setup, not a generic FAQ. Ask “why is Lexi not picking up emails?” and it pulls Lexi’s capabilities, recent run errors, and OAuth state, then proposes a specific fix you approve in one click. It also reviews your configuration on its own — nightly, and when you first open the widget — so misconfigurations surface before they bite. Owners set the cadence, or switch the automatic review off entirely, in Settings → Support Agent. Tier-locked features are surfaced as upgrade opportunities; genuine platform gaps still log straight to the roadmap.

Block. Tackle. Ship.

Free to start. Upgrade when your team does.

Start freeSign in
Featured on There's An AI For That
Start free →