On this page
Compare

How to choose an AI gateway
for agents that touch real systems.

Verosek is an AI and MCP gateway: keep your SDK, route model calls with fallback, set budgets and rate limits per key, connect your tools, and have each tool call decided by the key's rules before it runs. Moving takes two values, base URL and key.

By need

Which kind of tool do you need?

Verosek is for a platform or infrastructure team whose own agents write to production databases and APIs, and who has to show what an agent did. New to the words? Here is what a router, an AI gateway, an AI firewall and an MCP gateway each do.

You want one gateway for model calls and your agents' tool calls, with rules that read what each call does.

Verosek routes model calls, sets budgets and rate limits per key, connects your tools, decides each tool call before it runs, and seals a record of each request.

You want the widest provider list.

LiteLLM is an open-source gateway you host, with over 140 providers.

You pick a gateway by speed at high volume.

Bifrost is a high-performance open-source gateway across 23+ providers.

You want prompt tooling and a hosted control plane.

Portkey offers prompt templates, a playground and versioning, with a control plane it hosts.

You train and serve your own models.

TrueFoundry is an AI gateway for models, MCP and agents, on a platform that also trains and deploys models.

You run MCP servers in containers on a workstation.

Docker MCP Gateway runs each server in a restricted container, and its open-source gateway is built into its MCP Toolkit.

Questions

Questions to ask any AI gateway

Ask each vendor these, on your own queries and your own tools. Here is how Verosek answers.

Questions to ask any AI gateway
The questionVerosek's answer
Which providers does it route to, and what happens when one fails?Verosek routes to OpenAI, Anthropic, Google, Azure, Bedrock, Ollama and any OpenAI-compatible server. Calls fall back in priority order, with retries and cooldown after failures, and a residency rule is checked before any call.
Can each key carry its own budget and limits?Yes. Each key carries its allowed models, budget, rate limits, expiry and per-tool rules.
What does a tool rule read?The call itself. For a database tool Verosek runs, it reads the SQL statement itself, so a key can read one table and be refused a DELETE on any of them.
What happens to a tool with no rule?A tool Verosek runs is refused, and not offered to the model, until the key has a rule for it.
What happens when a check cannot run?Each key sets each check to enforce, flag or log, so you choose what stops a call and what only leaves a note. By default a blocking rule that cannot run refuses, and you can set one rule on a key to let calls through when the check itself breaks.
What happens when a tool call times out?Verosek tells the agent the tool timed out and treats the call as one that may have run: if a check you set to enforce then cannot run, the answer says which calls ran and tells the agent not to repeat them yet.
Who can check the record later?Give a regulator your public keys, and they can check each decision themselves. You can verify a trace through the API.
Where can it run, and on which plan?Enterprise: private cloud, VPC or air-gapped, deployed with us
What does it cost, and is the price published?Team is $999 a month, billed yearly. Enterprise has custom terms. Both plans start with a call.
Six tools

Six tools at a glance

Six tools at a glance
The pointVerosekLiteLLMPortkeyBifrostTrueFoundryDocker MCP Gateway
Built forOne gateway for model calls and tool calls, with rules that read each callRouting to over 140 providers, with spend controlsOne interface to over 3,000 modelsHigh-performance routing across 23+ providersAn AI gateway for models, MCP and agents, on a platform that also trains and deploys modelsRunning MCP servers in containers
Model routingOpenAI, Anthropic, Google, Azure, Bedrock, Ollama and any OpenAI-compatible server, with fallbackOver 140 providers, with fallbacks in open sourceOver 3,000 models, with fallback23+ providers, with failoverOver 1000 LLMs, with failoverOpen source: Running and proxying MCP servers in containers.
Tool rulesPer key and per tool, before each call runs: a SQL statement's operation, tables and row bounds, a payment's amount, a file path, a namespace, a recipientAllow lists at six levels, per key and team among them, with patterns on valuesTool switches and text and structure checksAllow lists per key; guardrails on arguments on EnterpriseCedar or OPA policies you writeOpen source: Allow lists by profile.AI Governance: Cedar rules on values.
ScanningPrompt injection, personal data and secrets in prompts, replies, tool arguments and tool results, on both plans, with no second vendor to buyA guardrail framework in open source; guardrails per key or team on EnterpriseDeterministic checks on Developer, LLM and partner checks on ProductionGuardrails on EnterpriseBuilt-in guardrails from ProOpen source: A scan for secret-like values.
The recordOne sealed record for each request, with the reason for each decisionRequest logs through callbacks; admin audit logs on EnterpriseLogs of each MCP request; Audit Logs of admin actions on EnterpriseRequest logs; signed audit logs of admin activity on EnterpriseMetadata audit logs of platform activity from ProOpen source: Tool names and argument shape.AI Governance: Metadata records, without parameter values.
Where it runsEnterprise: private cloud, VPC or air-gapped, deployed with usOpen source: self-hosted; hosted through salesEnterprise: managed SaaS or hybrid; open source: a self-hosted MIT gatewayOpen source: self-hosted; Enterprise: VPC, on-prem and air-gappedHosted; Enterprise: VPC and on-premOpen source: Any Docker Engine.Enterprise Gateway: Your own account, or air-gapped.

As of 5 Nov 2026.

The proof

Attacks become findings, and findings become rules on the key.

Assay, an open MCP security benchmark we maintain, was presented at Black Hat India Arsenal. Our red-team framework holds 205 attack techniques, 68 for MCP, with findings anyone can recompute.

Frequently Asked Questions

If you want one gateway for model calls and your agents' tool calls, with rules that read what each call does, talk to Verosek. Each key carries its allowed models, budget, rate limits, expiry and per-tool rules. For a database tool Verosek runs, it reads the SQL statement itself, so a key can read one table and be refused a DELETE on any of them. Model calls and tool calls land in one sealed record, with the reason for each decision.

With Verosek, no: the checks ship in the gateway, so there is no second vendor to buy, no second key and no second record. Turn on Shield and each key's checks look for prompt injection in prompts, tool arguments and tool results, and for personal data and secrets in those and in replies. If your gateway reaches a scanner through an integration, ask what that adds: an account, a key, a console, a bill, and where your text goes.

If your agents also call tools against production systems, add Verosek to the list: one gateway for both, with rules that read each call. LiteLLM has an open-source core you host, and Portkey's commercial product comes as managed SaaS or hybrid. As of 5 Nov 2026.

Verosek routes too, with rules that read each tool call on the same key. Bifrost sells on performance, and LiteLLM on reach, with over 140 providers. As of 5 Nov 2026.

Verosek puts model calls and tool calls on one key. Docker's open-source gateway runs MCP servers in containers, and TrueFoundry is an AI gateway for models, MCP and agents, on a platform that also trains and deploys models. As of 5 Nov 2026.

Verosek can. On the Enterprise plan it can run wholly in your environment, deployed with us: private cloud, VPC or air-gapped. Among the others, LiteLLM offers it on Enterprise, Bifrost on Enterprise, TrueFoundry on Enterprise and Docker on its Enterprise Gateway. Portkey has closed it to new customers. As of 5 Nov 2026.

Verosek is the gateway as well: one key routes your model calls, Bedrock among them, sets budgets and scans with no second vendor, and it reads the SQL statement and seals the reason for each decision. AgentCore Policy goes by user identity, a tool call's input parameters, earlier steps in the session and signals such as a guardrail's score; its docs do not say a policy reads a SQL statement. As of 5 Nov 2026.

Verosek is closed source. The gateway and its admin console hold no licence check, telemetry, analytics or call to a Verosek host.

Team is $999 a month, billed yearly. Enterprise has custom terms. Both plans start with a call. See the plans.

By default a blocking rule that cannot run refuses, and you can set one rule on a key to let calls through when the check itself breaks. A check that only flags lets the text through and names it in a response header. A request the record cannot hold is refused before anything runs.

An app leaves by changing the same two values. On a self-hosted deployment, on the Enterprise plan, the gateway, its signing key and your records stay with you.

Bring one real agent to a call, and we stand up a gateway for it.

Sources

Your agents touch real systems. Bring them to a call.

Still curious? Ask your favourite AI about Verosek.