the short answer
Self-hosting Future AGI means Docker Compose today; its docs say there are no Helm charts or Kubernetes manifests yet. The stack includes PostgreSQL, ClickHouse, Redis, Temporal, MinIO and a code executor (the README also lists RabbitMQ), needs Docker Engine 24.0+ and Compose v2.24+, and is sized from 4 cores and 8 GB of RAM for a trial to 16+ cores and 32+ GB for production. Telemetry is on until you set FUTURE_AGI_TELEMETRY_DISABLED=1.
- Install path
- Docker Compose only; Helm and Kubernetes manifests on the roadmap
- Production size
- 16+ cores, 32+ GB RAM, 200+ GB SSD, per the docs
- License
- Apache-2.0; code in ee/ directories needs a paid subscription in production
- Telemetry
- On by default; FUTURE_AGI_TELEMETRY_DISABLED=1 turns it off
The Requirements
Future AGI's self-hosting documentation gives three starting sizes:
| use | CPU | RAM | disk |
|---|---|---|---|
| Evaluation, a local trial | 4 cores | 8 GB | 20 GB |
| Team, 1-20 users | 8 cores | 16 GB | 50 GB |
| Production, 20+ users | 16+ cores | 32+ GB | 200+ GB SSD |
Software: Docker Engine 24.0 or later, Docker Compose v2.24 or later, and Git 2.0 or later. It runs on Linux bare metal, EC2, GCE, Azure VMs, and GKE or EKS nodes with privileged mode enabled. The docs rule out managed container platforms that block privileged containers - Fargate, Cloud Run, Render, Railway and Fly.io.
On Kubernetes, the docs are direct: "There are no Helm charts or Kubernetes manifests: that support is on the roadmap, and Docker Compose is the only supported path today." The GitHub README lists official manifests and Helm charts as coming soon.
What You Are Running
The requirements page lists PostgreSQL, ClickHouse, Redis, Temporal, MinIO and a code-executor service; the README describes the data layer as PostgreSQL for metadata, ClickHouse for spans and time-series, Redis for state, and RabbitMQ plus Temporal for jobs. The Docker Compose guide starts "all 21 services" in full-stack mode. Published ports include the frontend on 3000, the backend API on 8000, the gateway on 8090, model serving on 8080, the code executor on 8060 and OTLP collectors on 4317 and 4318.
The documented install is short:
git clone https://github.com/future-agi/future-agi.git
cd future-agi
cp .env.example .env # then replace every CHANGEME secret
docker compose upUpgrades are git pull, docker compose pull, docker compose up -d, and migrations run on startup. The operational weight is in what the Compose file hides: two databases, a cache, a workflow engine and object storage, each of which needs backups, disk monitoring and upgrades once the install matters.
License and Telemetry
Future AGI opened the platform under Apache-2.0 in Q2 2026. The repository also carries a LICENSE-EE file covering any directory named ee/, and it says that code "requires an active paid subscription (an Enterprise license key or a Future AGI Cloud plan) to use in production." If a feature you plan to use lives under ee/, budget for it or confirm with Future AGI.
Self-hosted instances send telemetry by default. Per the README, a registration on first boot sends the instance ID, version, deployment type, and the email addresses and domains of active admin users; a periodic heartbeat sends anonymous aggregate usage counts. Set FUTURE_AGI_TELEMETRY_DISABLED=1 in .env, or in deploy/.env.production for production, to turn it off. For an air-gapped or regulated install, make that part of the first boot rather than something you remember later.
Before Production
Future AGI's production guide opens with a warning worth taking at face value: "The default stack is not safe to expose. It boots with development secrets that work out of the box, no TLS." The guide pairs that with a security checklist, TLS termination, and backup and restore procedures for Postgres, ClickHouse and MinIO.
- Replace every
CHANGEMEsecret and disable telemetry before first boot if you need to. - Put TLS in front of the frontend, API and gateway ports.
- Schedule and test restores for Postgres, ClickHouse and MinIO, not only backups.
- Pin the image version and upgrade deliberately, since migrations run on startup.
What to Test During the Pilot
The repository's issue history points to the parts of a self-hosted deployment that deserve an explicit test before production:
- Fresh installation and upgrade: start from the published Compose file, ingest test data, then upgrade one release and confirm database migrations complete cleanly.
- End-to-end ingestion: send gateway requests and malformed spans, then verify valid traces still reach the dashboard and failed batches are visible.
- Authentication without public services: test signup and login with reCAPTCHA and other external dependencies disabled.
- Evaluator consistency: run the same deterministic check through the SDK and platform, then compare the stored results.
- Air-gapped operation: block outbound internet access and check the UI, model assets and document viewers for CDN dependencies.
Run these checks on the exact version you plan to deploy. A passing demo on a connected laptop does not prove that upgrades, ingestion and authentication will work in a restricted production network.
How Other Self-Hosted Options Compare
If the footprint is the obstacle, compare the services and support model of each option. Langfuse self-hosts with Docker Compose, Helm or Terraform templates and uses Postgres, ClickHouse, Redis and object storage. Latitude offers Docker Compose, Swarm, Helm and Railway deployments under MIT. Failproof AI separates its local policy runtime from its Cloud platform: the CLI runs policies without a server, while Enterprise self-hosting provides tracing, querying, evaluations, automated failure analysis and alerts on Kubernetes with ClickHouse and PostgreSQL. See self-hosted agent evaluation platforms for the full comparison.
When Not to Self-Host
If the reason for self-hosting is data residency rather than cost, Future AGI's enterprise page lists deployment into your own AWS, GCP or Azure account and a full air-gapped option, run with them. If the reason is cost, compare the machine and the on-call time against the managed plan, which starts free with 50 GB of storage and unlimited seats. A 16-core production box and the people to look after it are not free either.
FAQ
Can I run Future AGI on Kubernetes?
Not through an official Helm chart or Kubernetes manifest. The self-hosting docs describe Docker Compose as the supported deployment path and list Helm support on the roadmap. They do list GKE and EKS nodes with privileged mode enabled as hosts for the Compose stack.
How do I turn off Future AGI telemetry?
Set FUTURE_AGI_TELEMETRY_DISABLED=1 in .env, or in deploy/.env.production for production deployments. By default a self-hosted instance registers on first boot with its instance ID, version, deployment type and the email addresses and domains of active admin users, then sends anonymous aggregate usage counts.
Is everything in self-hosted Future AGI free?
The platform is Apache-2.0, but the repository also has a LICENSE-EE covering code in ee/ directories, which needs an Enterprise license key or a Future AGI Cloud plan to use in production. Check whether the features you need live there before you plan a free deployment.
Get Started
Failproof AI is free to start. It finds recurring failure modes across agent sessions using code-based and LLM-based evaluations, groups the evidence into findings, and recommends fixes. Bring the eval suite you already have, alert the right owner when behavior drifts, and turn a tested fix into a policy that prevents the failure from recurring. See pricing for the tiers.
Sources
Checked against each vendor's own site and docs on 2026-09-14. Products change; if a detail here is out of date, tell us at support@befailproof.ai.
- Future AGI docs: Self-hosting requirements
- Future AGI docs: Docker Compose
- Future AGI docs: Production deployment
- future-agi README
- future-agi LICENSE-EE
- Future AGI enterprise page
- Langfuse on GitHub
- Latitude docs: Deployment
- Failproof AI docs