Sovereign · Secure · Auditable

Deploy AI in your company without exposing your data.

The open‑source platform to automate your operations with isolated, audit-ready AI agents hosted wherever you want. Sovereign by construction.

Talk to an expert How it works

Licensed · Standards-based · Auditable

Apache 2.0 AFPS spec RFC 9457 OIDC SSE OpenAPI 3.1

Your company already uses AI. In chaos.

While you plan your next AI budget, these three problems are already running through your teams.

68 %

of your team already uses ChatGPT or Claude. From a personal account, with code or customer data pasted into the prompt.

Salesforce, 2025

8 / 10

agent POCs never reach production. Nobody knows who has access to what, credentials live in scattered .env files.

Gartner, 2026

× 3

your LLM bill every quarter. Zero attribution by team, by project, by tenant. Just a line item that keeps growing.

Field observation, 2026

Shadow IT, stalled POCs, opaque bills. Three symptoms, one cause: no infrastructure.

Before / after Appstrate.

Four concrete shifts, on the CIO and CFO side.

Without infrastructure

With Appstrate

6 months to wire a new agent into your systems.

AFPS manifest, deploys in minutes, portable across instances.

Prompts leaking to third-party clouds, often outside the EU.

Runtime hosted wherever you decide. Data stays in your perimeter.

No trace of who called which tool, with which permissions.

Audit trail per tenant, per scope, per run. OpenTelemetry streaming.

Opaque LLM bill, impossible to attribute.

Attribution per application, per end-user. Quotas per tenant.

Without infrastructure

6 months to wire a new agent into your systems.

With Appstrate

AFPS manifest, deploys in minutes, portable across instances.

Without infrastructure

Prompts leaking to third-party clouds, often outside the EU.

With Appstrate

Runtime hosted wherever you decide. Data stays in your perimeter.

Without infrastructure

No trace of who called which tool, with which permissions.

With Appstrate

Audit trail per tenant, per scope, per run. OpenTelemetry streaming.

Without infrastructure

Opaque LLM bill, impossible to attribute.

With Appstrate

Attribution per application, per end-user. Quotas per tenant.

Production AI. No compromises.

Sovereignty

Your data never leaves your perimeter. Self-host, dedicated EU cloud, or air-gapped. Your LLM keys stay yours (BYOK).

Governance

Every action traced. Every API key isolated. Every employee scoped. RBAC, audit trail, kill-switch in one click.

Measurable ROI

One agent covers around 20 % of a full-time role. Cost tracked per process, per team, per run.

Developer-first

Your internal devs own the workflows. AFPS manifests, CLI, git-reviewable. The webapp handles launch, scheduling and audit for everyone else. No low-code builder to outgrow.

The architecture that reassures your CISO.

Each run spins up a fresh container, destroyed after use. The agent reasons. The sidecar acts. Your API keys never see the LLM.

Audit trail · 1 run

// run_id=r_7a3f… tenant=acme-corp

[12:04:01]agent.callmodel=claude-4.6 keys=<none>

[12:04:02]sidecar.authprovider=gmail tenant=acme-corp

[12:04:02]sidecar.exchangetoken=ya29.*** ttl=55m

[12:04:03]tool.invokename=stripe.refund scope=refunds:write

[12:04:04]sidecar.redactfields=[email, pan]

[12:04:05]run.completeduration=4.2s events=14

→See the event schema

Runtime topology

Employee

Isolated container (ephemeral)

AI agent

never sees your API keys

Sidecar proxy

holds your credentials

Your systems

Ephemeral container

BYOK, keys in sidecar

LLM never sees secrets

Per-tenant isolation

Full audit trail

Air-gap capable

Measured on the platform. Reproducible.

Every number below links to its source. No claimed adoption figures.

Cold start

< 800 ms

pre-warmed sidecar pool

Benchmark

Sidecar overhead

< 12 ms

per tool call, local loopback

Benchmark

Container lifetime

Ephemeral

0 persistence cross-run

Docs

Isolation

Per-user · per-run

sandboxed container per run

Security paper

Permissions

78 typed

RBAC scopes across the API

RBAC reference

Audit events

Structured

OpenTelemetry streaming

Schema

Zero vendor. Your infra. Provable.

Same install as the docs. Same tiers. Same CLI.

~ appstrate install

$ curl -fsSL https://get.appstrate.dev | bash # or: bunx appstrate install

Pick a tier

Tier 0Bun · PGlite in-process · filesystemzero Docker

Tier 1Docker · PostgreSQL · filesystem

Tier 2Docker · PostgreSQL + Redis · filesystem

Tier 3Docker · PostgreSQL + Redis + MinIO · S3enterprise default

$ appstrate login # device-code flow, OS keyring, 15-min access token

✓ Sidecar proxy · 78 typed RBAC scopes · structured audit trail

✓ Ephemeral container per run · credentials never reach the agent process

✓ Zero telemetry by default. No phone-home. No license server.

Quickstart Progressive infrastructure Air-gap guide

Runs on: Raspberry Pi 4+ · laptop · VPS · bare metal · K8s · air-gap

Pi 4 (4 GB): Tier 0-1, 2-3 concurrent runs
Pi 5 (8 GB): Tier 2 comfortably, ~400 MB RAM at rest
Source: appstrate-oss/README.md, Bun ARM64 native

Source code: github.com/appstrate, Apache 2.0, auditable line by line.

Same binary, same CLI, from Tier 0 (dev) to Tier 3 (prod). No fork. No rewrite.

Compliance is a posture, not a checklist.

Architecture ready

Shipped, verifiable in code.

Data residency = wherever you self-host
Source auditable line by line (Apache 2.0) github.com/appstrate
78 typed RBAC scopes scopes reference
Structured audit trail, OTel streaming schema
Credentials never reach the agent (sidecar proxy) security docs
Ephemeral per-run containers, zero cross-run persistence
BYOK across 18+ LLM providers providers list
OIDC authentication auth docs

Compliance mapping

How our architecture serves common frameworks.

GDPR
Self-host in EU, audit trail, data you control.
EU AI Act
Audit trail satisfies logging obligations (Art. 12).
HIPAA
Sidecar isolation and audit trail give you technical safeguards baseline. Covered entity remains liable.
NIS 2 / DORA
Incident timeline reproducible from the audit trail.

We ship the architecture. You or your partner agency map it to your specific compliance obligations.

What we haven't done yet

Explicit roadmap, no made-up dates.

SOC 2 Type II · Not started. We will announce when engaged with an auditor.
ISO 27001 · Not started.
HIPAA BAA · Not offered. Self-host gives you the architecture; your DPA with your hosting provider covers the rest.
Standard DPA · Draft available on request, not yet templated.

Roadmap moves when customers with real compliance needs engage. Talk to us if SOC 2 or ISO 27001 blocks your procurement.

Talk to us about compliance

Honest comparison.

We lose on some cells on purpose. We win on what matters for sovereignty.

	Appstrate	Bedrock AgentCore	Claude Managed	Copilot Enterprise
Self-host, on-prem
Air-gap capable
Source code auditable
Apache 2.0
Multi-LLM (18+ providers)			Claude only	OpenAI only
BYOK (your keys)		AWS-bound
Multi-tenant end-user		org-level	org-level	org-level
EU data residency
Predictable pricing	Free + BYOK	per-hour + granular	per-hour + tokens	per-user per-month
Time to first run	5 min self-host	Immediate managed	Immediate managed	Immediate managed
Vendor lock-in	None (Apache)	AWS	Anthropic	Microsoft

Pratfall on purpose: we lose on "time to first run" (managed services are immediate, we take 5 min to deploy). We win on everything else that matters once you're in production.

Download comparison PDF (12 pages)

From decision to production. 6 to 8 weeks.

A methodology, not a client story. Same steps for every deployment.

Audit

Week 1-2

→ Sovereignty map

●Process map of your current AI usage
●Security audit with CISO and DPO
●ROI baseline measurement

Pilot

Week 3-5

→ First agent in production

●One process piloted end-to-end
●Self-host or EU cloud deployment
●Real ROI measurement on that process

Prod

Week 6-8

→ Tenant-ready runtime

●Full rollout with audit trail enabled
●Business team training (one morning)
●Operational handover to your teams

Extend

Month 3+

→ Internal agent catalog

●New agents from your internal library
●ROI measured continuously
●Cross-department extension

Audit

Week 1-2

→ Sovereignty map

●Process map of your current AI usage
●Security audit with CISO and DPO
●ROI baseline measurement

Pilot

Week 3-5

→ First agent in production

●One process piloted end-to-end
●Self-host or EU cloud deployment
●Real ROI measurement on that process

Prod

Week 6-8

→ Tenant-ready runtime

●Full rollout with audit trail enabled
●Business team training (one morning)
●Operational handover to your teams

Extend

Month 3+

→ Internal agent catalog

●New agents from your internal library
●ROI measured continuously
●Cross-department extension

Delivered by your team, or by one of our certified partner agencies.

Questions your committee will ask.

Legal / DPO

Where is our data hosted, and who has access?

You choose. Self-host (your servers, zero Appstrate access), dedicated EU cloud, or fully offline air-gap. In all three cases, data is never shared with OpenAI, Anthropic or Appstrate. Your LLM keys are yours (BYOK).

CISO / IT

What will our pentests and SOC actually validate?

Open‑source code under Apache 2.0, auditable line by line. Sidecar proxy architecture (the agent runs sandboxed without credentials). Ephemeral containers, zero cross-run persistence. RBAC with 78 permissions. Structured audit trail. Security docs and architecture diagrams delivered pre-sales.

CFO

How do we keep the AI bill under control?

Appstrate itself is free and open‑source. Your only direct costs are your infrastructure (VPS, servers) and your LLM tokens (BYOK, paid directly to the provider). Certified partner agencies handle installation and support, with pricing agreed directly with them.

HR / Business

Will our teams actually know how to use it?

Yes, in one morning. Business teams get a plain webapp: one-click launch, scheduling, run history, audit trail. Complex workflows are built once by your internal devs or a partner agency, then consumed self-service.

CEO / COO

How do I know this is not another POC that goes nowhere?

Pilot on one process in 5 weeks with a concrete ROI report. If the numbers do not hold, you walk away. Apache 2.0, no lock-in, no penalty. Our first three production clients all crossed pilot to production in under 8 weeks.

Is Appstrate right for you?

We'd rather tell you now than three months after the contract.

Yes, if...

You have a real sovereignty constraint (healthcare, defense, finance, public sector, EU data residency).
You have a platform team that can operate a runtime, or a trusted integrator.
You want multi-tenant infra, with per-end-user scoping and per-run audit trail.
You're ready to host (VPS, on-prem, EU cloud, air-gap).
You refuse single-US-cloud lock-in.

No, if...

You want a turnkey SaaS UI with zero infra work (Claude for Work or ChatGPT Enterprise will be faster).
No sovereignty or multi-tenant need (Bedrock AgentCore or Vertex will do).
You want a personal assistant (that's Claude.ai, not Appstrate).
You refuse any DevOps work at all (ask a partner agency).

Still unsure? The 30-min call settles it in 10.