Secure your agents.

The gofl platform gives agents the identity standards your security team already trusts. Connect any framework with one lightweight SDK.

Enterprise-grade governance

Govern every call

Every model, tool, MCP and database call passes one gateway — under one identity, one audit trail, one policy. Same governance, builder-built or your own code.

Sources · clients
CLI / IDE
AI agents
Apps & clients
Agent gateway · control plane
Agent Gateway
Identity & RBAC
Policy & guardrails
Secret injection
Rate limits & budgets
Routing & failover
Audit & tracing
Destinations · models & tools
ModelsOpenAIAnthropicGemini
MCP serversGithubNotion
DatabasesSnowflake
APIs / services
OTel collector
stream to your SIEM
Datadog Elastic Grafana+ your SIEM

Secrets never reach the model

The gateway injects a scoped, short-lived credential at egress. The agent and the LLM never see a standing secret.

Every call attributed

Bound to the signed-in person or a named service account — human or machine. Attributed, never NULL.

One audit trail

Model spend and every tool, MCP and database call in a turn share one trace under one identity.

Govern your own code

Declare models, tools and access as Kubernetes resources. Your engineers’ own agents get the same plane — as code.

No new security model

The standards we spent a decade building — now for agents

OAuth and OIDC gave enterprise software scopes, resource owners and resource servers. Agents arrived and threw all of it out. gofl brings those same primitives to every agent call — so your security team governs agents with the model they already trust, not a bespoke one.

scope

Bounded permission, per call

An agent can only do what its scope allows — never the user’s full access. The same idea that has bounded API clients for years, now bounding agents.

resource owner

The person the agent acts for

Every call is tied to the human — or service — on whose behalf it runs. The resource owner is explicit, not lost the moment an agent takes over.

resource server

Your models, tools, DBs and APIs

Each destination validates the token and enforces its own access. The gateway never bypasses it — exactly how resource servers have always worked.

delegation · token exchange

On-behalf-of, provably

The agent presents a short-lived token that names both who acted and on whose authority — standard delegation, applied to autonomous agents.

gofl SDK

Lightweight SDK

The whole integration is a base URL and an import. Reach any model through your gofl key, and call any tool on a user’s behalf with one GoflAuth — in the framework you already use.

Models
gofl_llm.py
1from openai import OpenAI
2
3# Your gofl vkey — created once, then it's just an env var.
4llm = OpenAI(
5 base_url="https://app.gofl.io/v1",
6 api_key=os.environ["GOFL_VKEY"], # ← any model, one key
7)
8
9resp = llm.chat.completions.create(
10 model="gpt-4o",
11 messages=[{"role": "user", "content": "Summarize #4471"}],
12)
Tools · MCP
gofl_mcp.py
1from gofl_identity import GoflAuth
2
3# One GoflAuth, reused for every request.
4gofl = GoflAuth(
5 issuer=os.environ["GOFL_ISSUER"],
6 client_id=os.environ["GOFL_AGENT_CLIENT_ID"],
7 client_secret=os.environ["GOFL_AGENT_SECRET"],
8)
9
10def mcp_client():
11 from langchain_mcp_adapters.client import MultiServerMCPClient
12
13 return MultiServerMCPClient({
14 "gofl": {
15 "transport": "streamable_http",
16 "url": os.environ["GOFL_MCP_URL"],
17 "auth": gofl, # ← the whole integration
18 }
19 })
Drops into
OpenAI OpenAI SDKLangChain LangChainLangGraph LangGraphGoogle Google ADKCrewAI CrewAI Any httpx client
Open by design

Any model, any provider

Open-weight models on your own GPUs, or a frontier model reached through the provider you already run. One contract for all of them — swap the model, not your code.

  • Sovereign, on your hardware

    Run open-weight models on your own GPUs — in-region, nothing leaving your boundary.

  • Frontier, through your provider

    Reach the latest hosted models through the cloud account you already run. The provider credential never leaves the gateway.

  • One model contract

    Every model — self-hosted or hosted — speaks the same API. Switching is a string.

Reached through
Bedrock BedrockVertexAI Vertex AIAzureAI AzureGroq Groqtogether.ai Together self-hosted vLLM
Gemini
Mistral
OpenAI
Meta
Anthropic
DeepSeek
Deploy anywhere

Deploy where you are comfortable

Self-host on the infrastructure you already run, or let us run it for you — the same platform and the same governance, either way.

gofl installs as an appliance on the infrastructure you already run — any cloud, your own Kubernetes, bare metal, even air-gapped. Nothing phones home. One helm install, one knob: a domain.

terminal
$ helm install gofl gofl/gofl \
      --set domain=gofl.acme.com
Runs onsame appliance, everywhere
AWSAWS
AzureAzure
GoogleCloudGoogle Cloud
Kubernetes
OpenShift
Bare metal
On-prem
Air-gapped

Book a demo.

Tell us what you want to build and we’ll get back to you.

Or email me directly — [email protected]

By submitting, you agree that Conamur Software AB will use your details to contact you about a demo. See our Privacy Policy.