The hub
SpanDock
Stores everything, forwards it, manages clients and their settings, and offers MCP and a built-in assistant. It signs in to spandock.com once to activate.
- Self-hosted
- Air-gap ready
- 0 inbound ports · WireGuard
- Telemetry stays on your machines
The control plane for AI coding tools
Claude Code, Codex, Gemini CLI, Copilot and more report to a SpanDock server you run. Set budgets, model routing, guardrails and MCP permissions once, and every computer follows them. You see every session, tool call and dollar, and prompts and tool contents stay redacted at the source on every plan.
- AI tools supported
- 13+
- AI tools supported
- MCP connectors
- 142
- MCP connectors
- backends to forward to
- 16
- backends to forward to
- starter skills
- 36
- starter skills
- open ports
- 0
- open ports
Source: SpanDock docs and integrations catalogue, updated 2 November 2026
curl -fsSL https://spandock.com/deploy/install-server.sh | sudo bashRunning in under a minutemacOS · Linux · Windows · no Docker
spandock · linux serverOne command
What's real: the ==> lines are the install script's own messages, abridged. The activation link is an example.
Live
Today · guardrails
Kept out of prompts
- Blocked
- 7
- prompts and tool calls
- Redacted
- 12
- tool results
- Warned
- 3
- flagged for review
By detector
- Secrets9
- Email addresses8
- Card numbers5
| Event | Detector | Action |
|---|---|---|
| Claude Codeprompt · checkout-api | Secrets | Blocked |
| Codextool result · crm search | Email addresses | Redacted |
| Gemini CLItool call · billing | Card numbers | Blocked |
| Claude Codeprompt · infra | Words | Warned |
Live traffic
Watch a request go through SpanDock
Every AI tool call starts on a computer, where guardrails check it, then passes a satellite to your hub, which applies budgets, routing and MCP rules before it reaches a model or an MCP server.
What this shows: a simulated SpanDock deployment with three computers, two satellites and one hub. Guardrails on each computer stop calls that contain a secret or a card number before they leave it. The hub moves a person's calls to a cheaper model when they near their daily budget, applies routing rules, denies destructive MCP tools, and records telemetry for its dashboard and for forwarding. The counters, savings and latest calls below are part of the simulation.
Live traffic
Hub · 2 satellites · 3 clients Simulated demo data
- Requests
- 15
- 2 MCP
- Blocked on computers
- 2
- guardrails
- Blocked at the hub
- 0
- budget · MCP rules
- Rerouted
- 0
- budget · rules
- Spend
- $1.34
- 47.1k tokens
- Saved
- $0.54
- vs the asked-for model
RequestResponse · tokens · costMCP tool callCheaper model (saving)BlockedThe hub
Savings
The asked-for model's price minus the answering model's.
$0.54
Budget reroutes$0.00
Routing rules$0.00
Blocked calls not paid for (estimate)$0.54
Routing and budgets
Dana's daily budget. At 80% her calls move to a cheaper model.
76%
Within budget$1.13 of $1.50 today
Latest calls
The six newest.
| Call | Model | Cost | Result |
|---|---|---|---|
| OpenCode · AlonOpenCode → hub → OpenRouter | router-model | $0.055 | Allowed |
| Qwen Code · CI runnernotion.search | MCP tool | $0 | Allowed |
| Qwen Code · CI runnerQwen Code → hub → Anthropic | claude-sonnet | $0.075 | Allowed |
| Gemini CLI · CI runnerGemini CLI → hub → Bedrock | bedrock-model | $0.184 | Allowed |
| Codex · DanaCodex → hub → OpenAI | gpt-mini | $0.033 | Allowed |
| Codex · Danagithub.get_issue | MCP tool | $0 | Allowed |
Simulated data. Guardrails run on each computer (Enterprise, and in the 30-day trial). Each client connects over WireGuard to its assigned server; satellites take their settings from the hub and send every record to it. Budgets, routing and MCP permissions are decided by the server. Simplified: every model call is drawn through the hub, and only API-key traffic goes through the model proxy. Model names and prices are placeholders.
Secure by default
Read the Trust pageSecure by default. Air-gapped when you need it.
Your server, your disks, your network: SpanDock needs no inbound port, keeps working offline, and with an Enterprise offline license never contacts us.
- Self-hosted
- Air-gap ready
- 0 inbound ports · WireGuard
- Telemetry stays on your machines
Self-hosted
SpanDock is a single app with an embedded database on your own machine. Telemetry leaves your network only for a destination you add, and it needs no Docker and no cloud account.
No inbound ports
Clients reach their server over an encrypted WireGuard tunnel, so a cloud VM needs no inbound port and no public address. The dashboard listens only on the machine itself.
Air-gapped on Enterprise
Start a server with
-license-file FILEor runspandock license import FILE. License keys are signed with Ed25519, so a server verifies its license offline, with no phone-home for licensing.Keeps working offline
Clients queue uploads on disk while their server is unreachable, and computers keep enforcing their guardrails. An online license has a 14-day grace period if the site can't be reached.
Checked on the computer
Guardrail matching runs on each computer, so prompt text never leaves it to be checked (Enterprise, and in the 30-day trial). Prompts and tool contents are redacted at the source on every plan.
What reaches spandock.com
With an online license, a license check sends a random installation ID, the role, version, OS and counts of clients and servers: never telemetry, prompts or code. With an offline license, nothing is sent for licensing.
| Deployment | For | Contact with spandock.com |
|---|---|---|
| Standard self-hosted | Most teams | One-time activation, then a license check once a day |
| Network-isolated | Hardened networks | Egress can stay blocked until a refresh; the server runs on its last license through the grace period |
| Air-gapped (Enterprise) | No route to the internet | An offline license file: no contact for licensing |
Control
See connectors and skillsSet the rules once. Every computer follows them.
Your server holds the rules for connectors, guardrails, skills and access, and each computer picks them up within a minute. Prompts and tool contents are still redacted at the source, on every plan.
MCP gateway
Your organisation's MCP connectors in every AI tool, through one gateway. Each starts read-only, destructive tools wait for a person to approve them, and permissions follow people and groups.
Premium and EnterpriseHow it works: MCP gateway
Guardrails on computers
Secrets, email addresses, card numbers, IBANs, national ID numbers and your own patterns are caught on each computer before they leave. Block, warn or ask the person to confirm.
EnterpriseHow it works: Guardrails on computers
Shared skills
Publish your team's own skills once, and every computer's AI tools get them, kept up to date. 36 starter skills for engineering, sales, finance, legal and more are ready to turn on.
Starter skills: every plan. Publishing your own: Premium and Enterprise, in betaHow it works: Shared skills
Just-in-time access
No standing write access to sensitive services: people ask for it for a while, with a reason and a ticket, and an approver grants it. The grant ends by itself.
EnterpriseHow it works: Just-in-time access
Change management
A report of what changed in production through AI tools, who changed it, what approved it and the ticket, with quarterly access reviews, exported for auditors.
EnterpriseHow it works: Change management
Notifications
A computer waiting for approval, a satellite offline, a disk nearly full: your server tells you on your spandock.com dashboard. Names and counts only, never telemetry.
Every planHow it works: Notifications
Security
Read the security overviewFind the risks on every computer, without collecting code.
Scans and checks run on each computer and on your servers. What reaches the hub is counts, advisories and fingerprints, never source code or a secret's value. Telemetry stays on your machines and goes only where you send it.
Repository security
One policy for the dependency and secret scans on every computer, with git hooks and CI that enforce it, and fleet reports. Secrets are reported as fingerprints; source code and secret values never leave the computer.
EnterpriseHow it works: Repository security
Process inventory
See which AI tools run on each computer, including the ones that send no telemetry. Never command lines or environment variables.
EnterpriseHow it works: Process inventory
The exposure check
The spandock exposure command checks an installation as an attacker would: on the machine itself, from another computer in your organisation, and across a private network range. Each finding comes with its fix, and the check never changes anything.
Every planHow it works: The exposure check
SSO and the audit log
Sign in to the dashboard with your identity provider, and keep a record of every administrative change: who made it, what changed, when and from where.
EnterpriseHow it works: SSO and the audit log
Cost
AI tools are a line item. Treat them like one.
With the model proxy on, your server decides which model answers each call and records what it cost, per person, project and model. The keys stay on the server.
Per-person budgets
A daily, weekly or monthly limit for each person. Near it, only cheaper models are used; over it, calls are blocked, sent to the cheapest model or flagged. SpanDock Mind, the advisor, picks a cheaper model for a conversation when it's good enough.
Standard and upHow it works: Per-person budgets
Routing rules
Swap models with rules, for example Opus to Sonnet, with retries and a backup provider. A conversation keeps its model for 6 hours, so it doesn't change mid-task.
Premium and EnterpriseHow it works: Routing rules
Token Saver
A local proxy on each computer that compacts only the newest content in a request (JSON, logs, test output, diffs), so the prompt cache stays valid. A dry run shows what it would have saved first, and the server has a kill switch.
Every plan, in betaHow it works: Token Saver
The model leaderboard
Rank the models your tools used by spend, tokens, calls, speed, reliability or cost per million tokens.
Every planHow it works: The model leaderboard
Visibility
Read the docsSee what your AI coding tools actually do.
Every session, tool call, model call, token and dollar, per person, project and model, stored on your own machines. Tool profiles keep prompts and tool contents redacted.
Sessions per project
Open any session to see every call in order, with its model, tokens, cost and duration.
Every planHow it works: Sessions per project
Delivery reporting
Commits, AI-assisted share, reverts and AI spend per commit, by team and repository. Never code, commit messages or email addresses.
Premium and EnterpriseHow it works: Delivery reporting
AI enablement report
Who uses AI coding tools and how much, what it costs and saves, and which teams lag, compared with the period before.
Premium and EnterpriseHow it works: AI enablement report
Project memory
Your team's AI tools fetch a current summary of the project (recent work, decisions, open threads) with /spandock.
Premium and EnterpriseHow it works: Project memory
Forward to your backends
Send telemetry to Langfuse, Grafana Cloud, New Relic, Datadog, Honeycomb and 11 more backends, or any OTLP endpoint, with a copy kept on your server.
One destination on Personal; any mix on Standard and upHow it works: Forward to your backends
Ask in plain English
Ask the built-in assistant about your telemetry, or let chat tools query it over MCP with read-only tools.
Every planHow it works: Ask in plain English
Works with your AI tools
- Claude Code
- Codex
- Gemini CLI
- GitHub Copilot
- Qwen Code
- Goose
- Mistral Vibe
- OpenCode
- Cline
- Kilo Code
Forwards to your backends
- Langfuse
- Grafana Cloud
- New Relic
- Datadog
- Honeycomb
- Dynatrace
- Elastic
- Splunk
- OpenTelemetry
Runs anywhere
One app, four ways to run it
Install the same app on Mac, Linux or Windows. On first launch, each computer becomes a server, a satellite or a client, and remembers it.
Scale out
SpanDock Satellite
Extra servers that join a hub to spread clients and delivery across machines. Clients fail over between servers automatically. Premium and Enterprise.
Every laptop
SpanDock Client
Reports to a server over an encrypted WireGuard tunnel, with no account and no open ports. Its settings can be managed from the server, so nobody has to touch it.
Your own computer
Just you
Run a server on your own machine: dashboard, storage and forwarding in one place. Sign in once to activate it; Personal is free.
Live failover
Clients move to another server by themselves
Each client gets a primary server and an ordered fallback list. When a satellite goes away, its clients reconnect to the next server within seconds and replay anything they queued. No one has to touch a laptop.
- Sticky by default: clients only move when they must.
- Fallbacks are spread, so one failure never overloads one server.
- A live cluster diagram shows where every client is connected.
All servers online. Each client sends to its primary server.
Pair many computers at once
Import people from a CSV file, add the computers you expect, and make a single-use pairing code for each of them at once, to hand out or push with your device management.
Settings managed from the server
Set a client's settings on the server, one computer or many at once. A setting you set is read-only on the client; clients apply changes within a minute and report back.
Uptime on your dashboard
Your servers' uptime on your spandock.com dashboard, from the last 24 hours to 90 days, with each satellite's. The heartbeat carries no names, addresses or data.
Every planHow it works: Uptime on your dashboard
Backups to your own storage
Scheduled, sealed backups of the server to your own object storage, and a restore onto a new machine.
Standard and upHow it works: Backups to your own storage
A live cluster view
Watch servers, satellites and clients on one diagram, with live traffic flowing between them.
Say hello
Meet Spander and Byte
Spander is our site guide. He waits at the top of every page. Ask him anything about SpanDock: he finds the answer in the docs, or passes your question to the team.
Byte is his robot-dog sidekick, the face of the watchdog that keeps an eye on your server. He's a drawing, so he never tells you whether your server is up: your dashboard does.
Pricing
Compare plansStart free. Pay for what you run.
Every server needs a license, and Personal is free: one server and up to 3 users. Upgrade for unlimited users, teams or satellites.
Personal
$0
forever
- One server, up to 3 users
- Sessions, cost, model proxy and assistant
- One destination
Standard
$6
per active user / month
- Sessions, cost and the model proxy
- Per-person budgets and the model advisor
- One server, unlimited users
Premium
$400
per month
25 users included, then $4 per user
- Everything in Standard
- Five teams, shared keys, rate limits
- Routing rules, satellites and IP allowlisting
Enterprise
$1,200
per month
100 users included, then $3 per user
- SSO, SCIM and audit log
- Secret storage and per-team guardrails
- Offline license and weekday support