Implement · Berget AI
Can you run it?
Ownership levelSubstantialnone·limited·partial·substantial·fullAnalytical input C ยท 67.2/100
This page is a projection of the one entry record, the Reliability and Data control factors that Implement covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.
Which domain expands which factor
- AssessUse & modify + Transparency
- ImplementData control + Reliability
- UseReliability
- SupportTransparency
API integration
Berget exposes an OpenAI-compatible chat/completions API at https://api.berget.ai/v1 -
point an existing OpenAI client at the base URL and API key. Because the surface is
OpenAI-shaped, it works with LangChain, LlamaIndex and the standard OpenAI SDKs, and the Mastra
model directory lists it as an OpenAI-compatible provider. Integration is a configuration
change. The API listing (GET /v1/models) also reports each model's capabilities, lifecycle
status and price.
Authentication & account setup
Authentication uses bearer API keys. Migrating from OpenAI means switching the API key and
endpoint. Fine-grained organisation/project key scoping and rotation policy are not documented;
if key-scoping is a control requirement, confirm the available granularity in the account
console before rollout. API access needs an active plan, and the free Trial needs a card at
signup.
Region & ZDR configuration
There is no region toggle for serverless inference. DPA Sec 10.1 lets the customer choose a
data centre for compute and storage only, "not including serverless inference", and all Berget
data centres are in the EEA. Zero data retention is the default for serverless inference under
DPA Appendix 2 and ToS Sec 7, so no sales step or setting is needed to get it. A named data
centre or reserved capacity is a Dedicated Inference or cluster conversation with sales, and
there is no API flag for it.
Portability & exit
The exit path is clean. Berget speaks the OpenAI-compatible API over open-weight
checkpoints (Gemma 4, Mistral-Small, GLM-5.3-Flash, Kimi K3.1 Pro), so the same models run on
other providers or self-hosted stacks. Switching cost is dominated by re-pointing the base URL
and re-validating outputs. On exit, the Terms give a 30-day post-termination export window
for Customer Data before irreversible deletion. The catalogue turns over: Llama, gpt-oss-120B and
GLM-4.7 were retired between June and September 2026, with 2 to 3 weeks notice for Active models
and none for Preview models. Pin to checkpoints you can run elsewhere.
How this scores
The ownership factors this domain covers, drawn from the one entry record.
3
ReliabilityDoes it stay up and stay secure?
ModerateServerless inference is best effort with no compensation under the SLA; only sales-contracted Dedicated Inference with an HA setup carries 99.5% with credits. The status page has no readable content, and there is no independent pen-test or attestation.
How this scores (AOI sub-dimensions)
Reliability2/5whether it stays up, with an SLA and status historyA published SLA (2026-10-23) gives serverless inference best effort with no compensation.
Security3/5the controls protecting your traffic and dataToS Sec 6.7 states encryption in transit, network isolation, access controls and system monitoring.
Compliance2/5which independent certifications and attestations it holdsA GDPR DPA is published (Art.
4
Doesn't extract your dataDo the binding terms keep your data and IP yours?
StrongThe DPA states true Zero Data Retention with no persisted disk during inference and an express no-training sentence. The ToS keep title and IP with the customer under a licence scoped to providing the Service, and Customer Data is exportable for 30 days after termination then deleted. All data centres are in the EEA and Berget has no US parent. Pinning is not available for serverless inference.
How this scores (AOI sub-dimensions)
Data governance4/5retention, training-on-inputs and data ownershipThe documents give zero data retention for serverless inference (DPA Appendix 2: no storage after completed calls and no persisted disk even during inference; ToS Sec 7) and an express no-training sentence in the DPA (Appendix 2).
Residency4/5where your data is processed and storedThe binding position is an EEA boundary: all Berget data centres are in the EEA and no transfer leaves the EEA without the customer's written instruction (DPA Sec 10.1, 10.2), and Swedish ownership puts Berget outside US CLOUD Act reach.
What this means for adoptionYou substantially own workloads here. The binding DPA and Terms state zero retention of prompts and outputs, an express no-training sentence, customer-retained ownership under a service-scoped licence, a mutual confidentiality clause and a 30-day export window. An EEA-only data-centre boundary and Swedish ownership keep you outside US CLOUD Act reach, and a portable open-weight exit keeps switching cheap. Verification depth and operating terms hold it below full ownership. No independent audit or SOC 2 or ISO 27001 attests the guarantees, and serverless inference is best effort with no SLA credits. The contract promises EEA data centres with no data-centre choice for serverless, and Sweden-only is marketing. Adopt as an EU-based default for non-critical availability, and contractualise Dedicated Inference if you need an uptime commitment.
Sources
The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.
Terms of serviceread2026-11-19
Berget AI Terms of Service (page stamp 2026-10-16).
Privacy Policyread2026-11-19
Berget AI Data Privacy Policy (last updated 2025-03-10): a website privacy policy for visitors, listing website data, cookies and website retention periods.
Securityread2026-11-19
Berget AI Security page: a "Responsible Disclosure" policy with GitHub private vulnerability reporting, security@berget.ai with a PGP key and a Signal number, a typical 48-hour response, a scope list and a safe-harbour statement.
Data Processing Addendumread2026-11-19
Berget AI Data Processing Agreement (last updated 2026-10-22), an Art.
Slaread2026-11-19
Berget AI Service Level Agreement (last updated 2026-10-23), part of the Agreement under ToS Sec 3.1.
Vendor announcementread2026-11-19
Berget AI landing page (marketing): "Open models running under Swedish jurisdiction with zero data retention", "never stores the content of your prompts or generated output, and never trains on your data", "EU-only infrastructure, no US entity involvement".
Documentationread2026-11-19
Berget AI developer documentation (including llms.txt and Markdown pages): OpenAI-compatible chat completions API at https://api.berget.ai/v1 and a model lifecycle policy (Preview models removable without notice; Active models retired with 2 to 3 weeks notice and a named successor).
Documentationread2026-11-19
Berget AI live model listing (GET https://api.berget.ai/v1/models): 15 entries, priced in EUR per million tokens.
Third-party analysisread2026-11-19
Independent Mastra model directory lists Berget.AI as a provider reached through the OpenAI-compatible /chat/completions endpoint at https://api.berget.ai/v1, with 6 models routable.
Third-party analysisread2026-11-19
Independent EU tech press (Tech.eu, 2026-02-02): Berget AI raised EUR 2.1 million (SEK 24 million) in a round led by Luminar Ventures with Wellstreet and Norrsken Evolve.
Vendor announcementread2026-11-19
Berget AI pricing page (the per-model table is JavaScript-rendered, so rates come from ev-catalogue).
Vendor announcementread2026-11-19
Berget Code launch post (2026-05-13): agentic coding assistants.
Vendor announcementread2026-11-19
Berget AI Dedicated Inference page: reserved GPUs in Swedish data centres for a chosen or customer-supplied model, sales-led ("Talk to us").
Slaunverified2026-11-19
Berget AI status page.