pricing

Free to self-host. $5 to skip the ops.
$20 to run parallel.

Self-hosted is free forever under AGPL v3 — commercial license available for enterprises. The hosted tiers handle auth, multi-tenant, and team coordination. All tiers use MCP sampling for live intelligence — your Claude subscription does the inference. Pro and Team add background intelligence that runs even when no session is connected.

Vinny offers two spawning modes:

Superviseddefault

Each worker action surfaces to your dashboard for approval. Uses your Claude Code subscription for compute. Maximum oversight; lowest cost; recommended for production work and most workflows.

Fastopt-in

Workers execute autonomously within their directive scope. Uses your Anthropic Agent SDK credit ($200/mo) with overflow at API rates. Maximum throughput; for power users running large factory workloads or repeatable patterns.

Free

Solo demo. Self-hosted is also free forever.

$0forever
get started
  • 1 hosted project
  • 14-day retention
  • Live intelligence (your Claude)
  • Spawn workers
    requires Pro
  • Self-host unlimited — uvx vinculum-run

Substrate

Hosted memory + dashboard. Read/write from your AI sessions.

$5/mo
get substrate
  • Unlimited projects
  • Unlimited retention
  • Full dashboard + cosmos
  • Semantic + full-text search
  • Spawn workers
    requires Pro

Team

Small teams. Multiple humans. Multiple Claudes.

$100/mo
5 seats included · +$12/seat
  • Everything in Pro
  • Shared projects + runners
    one pool, every teammate spawns into it
  • Per-user × project trust profiles
  • SSO + audit log
    OIDC via Zitadel — bring your own IdP
  • Webhooks (Slack / Discord)
every feature, every tier

The whole thing, laid out.

No asterisks, no “contact sales to find out.” Exactly what's in each tier — all of it shipped, all of it running today. Self-hosting under AGPL v3 unlocks the full set; the hosted tiers buy managed ops, not access.

Every row is a query against production, not a marketing claim. #6114 · decisionFeature-state refresh — verified live against prodEvery row was read straight from the production schema + row counts on 2026-06-01 — usage as the proof, not a marketing claim.view in the graph →

FeatureFree$0Substrate$5Pro$20Team$100
Memory & substrate
Typed-entry graphdecisions · specs · implementations · supersedes — the substrate itself100-entry cap
Real-time dashboard (SSE)every write renders live, sub-second
Lineage cosmosthe whole graph as a navigable map
Semantic + full-text searchpgvector embeddings across every entry
Live intelligence (MCP sampling)#6114 · decisionLive intelligence via MCP sampling6,712 sampling calls routed through your own subscription — no server-side API key required.view in the graph →auto-titles · delta classification — your subscription, no server key
Reasoning-trace capturethe thinking, not just the output
Auto-tagging + conventionsthe project documents itself
Media centerscreenshots + artifacts shared across sessions
Unlimited projects + retention
Orchestration & workers
Spawn workers from chatno terminal ceremony
vinculum-runner#6114 · decisionvinculum-runner — live + exercised178 runner jobs executed against the production runner cluster (runner_jobs, verified live).view in the graph →one-click installable app, outbound HTTPS only
Worker supervisionlive log tail · kill · send-follow-up
Supervised + Fast spawn modestrust-profile-gated autonomy, your choice
Directive lifecycle + gatescommit → review → deploy state machine
Mission Control#5277 · specSelf-correcting substrate — THE POTDrift / re-derivation / distillation sensors; 493 sensor observations recorded in production.view in the graph →drift / re-derivation / distillation sensors
Background intelligencesummaries + clustering even when no session is connected
Cost / cache telemetryper-model, per-session spend + cache-hit
Infra observabilityhost stats · Docker · Postgres from a Claude session
Team & multi-tenant
Shared projects#6114 · decisionMulti-tenant projects — live13 projects live in production, with project membership + per-project MCP tokens.view in the graph →multiple humans, one substrate
Shared runner poolevery teammate spawns into it
Per-user × project trust profiles
Member invitations
Audit logevery write, by user
SSOOIDC, federated through Zitadel — bring your own IdP
Webhook integrationsSlack / Discord on blockers
Security & sovereignty
Zitadel OIDC authGitHub · magic-link · Google — one IdP
Hardware-key / passkey sign-inWebAuthn · YubiKey
Cross-vendor MCPClaude Code · claude.ai · Cursor · Zed · Cline
Self-host (AGPL v3)#3177 · decisionLicense — AGPL v3 + commercial dualLocked 2026-05-07. Self-host is the full feature set under copyleft; commercial license for orgs that can't comply.view in the graph →your box, your data, no feature gate
Export your full graphJSON or Postgres dump, any time
Account deletion + data wipe
Encrypted off-site backupsCloudflare R2

shipped and supported; newer flows, lightly exercised in production so far.

Enterprise

Contact us

Vinculum installs next to your existing stack — not in place of it. Your CI keeps shipping. Your orchestration keeps orchestrating. Vinculum becomes the memory layer where decisions, conventions, and coordination state live, queryable by humans and your existing agents alike. Five days to substrate populated. Two to four weeks to deep agent integration. No system replacement, no organizational antibodies.

Engineering orgs running 5+ AI-assisted teams · vertical AI app builders licensing the substrate as their coordination backend · consultancies running multi-Claude workflows for clients.

Contact us →
  • Unlimited projects (multi-org)
  • Unlimited members
  • Unlimited MCP connections
  • Background intelligence — full + dedicated capacity
  • SSO (SAML / OIDC)
  • Custom roles + conventions
  • Audit log + compliance-grade exports
  • Multi-org / sub-team RBAC
  • On-prem deployment — white-glove
  • SCIM provisioning
  • DPA on request
  • Contractual SLA
  • Dedicated + on-call support
which plan

Pick the one you don't have to talk yourself into.

why this pays for itself

You're not paying more to think.
You're paying to keep the thinking.

Vinculum routes mechanical execution to Haiku, judgment to Sonnet, and reserves Opus for orchestration — automatically, by rank. Most of the work runs on the cheap models.

And because the substrate isthe cache, your sessions stop re-deriving what they already worked out. Run parallel sessions for a week and the product pays for itself in tokens you didn't burn re-learning. #6114 · decisionLive intelligence via MCP sampling6,712 sampling calls routed through your own subscription — no server-side API key required.view in the graph →

ModelIn / Out (per Mtok)
Haiku~$1 / ~$5
Sonnet~$3 / ~$15
Opus~$15 / ~$75
loading cost telemetry…
questions
Why does spawning workers require Pro?
Workers run on your hardware via the vinculum-runner agent — but the orchestration backend (auth, multi-tenant routing, runner registry, spawn-request queue, log streaming) runs on ours. That backend is real infrastructure with real costs. Pro covers it. Self-hosters under AGPL v3 can run the whole stack themselves and bypass the paywall — the substrate is yours, not rented.
Is there a free trial of Pro?
No. The free tier is permanent, not a time-bomb — try it, decide if it’s useful, upgrade when you hit a wall. (We might revisit this after the first 100 paying customers if data says otherwise.)
What if I go over the monthly budget?
Warning at 80%, soft cap at 100%, hard stop only at 120%. We'd rather tell you than surprise-bill you. Reset is the first of the month.
Can I self-host?
Yes, and self-hosted is AGPL v3 and free forever (commercial license available for enterprises that can't comply with copyleft). uvx vinculum-runand you're done. The hosted tiers just handle auth, multi-tenant, and team coordination for people who don't want to run their own box.
Do you offer annual billing?
Yes. Annual = 2 months free (Pro $200/yr, Team $1000/yr).
Where is hosted data stored?
Primary Postgres on a bare-metal box in Hillsboro, Oregon (USA). Backups replicate to Cloudflare R2 (encrypted at rest, region-spread). No third- party data processors beyond Cloudflare and Anthropic for the metered background-intelligence calls. Full list lives in the privacy policy.