What does an AI gateway cost to build, partner or buy as SaaS?
The cost of an AI gateway depends mainly on logged request volume, how long prompt logs are kept, whether you need SSO and audit logs at enterprise level, and how much day-to-day operations work your team keeps. There are three realistic routes: run an open-source gateway such as LiteLLM or Apache APISIX yourself, have a partner implement and operate it in your infrastructure, or buy a hosted gateway plan. DeployOpen does not publish a price list. A quote follows from written scope, so the first useful step is to write down your expected request volume, retention period, identity requirements and who will run the platform.
The self-hosted route has costs that recur even when the software is free. The LiteLLM production guide names PostgreSQL as the primary database and requires Redis 7.0 or newer once more than one proxy instance runs, so you pay for database and cache hosting, backups and monitoring in your own accounts. Enterprise features are a separate line item: LiteLLM's SSO is free for up to 5 users from v1.76.0, and SCIM, audit logs, organization and team RBAC and IP allowlisting sit in the Enterprise license, which you should confirm with the vendor before budgeting. Ask a partner to separate one-time integration work from the monthly cost of operating, upgrading and supporting the gateway.
The hosted route is priced differently. Portkey's pricing page lists a free Developer plan with 10k recorded logs a month and 3-day log retention, a Production plan at $49 a month with 100k recorded logs and 30-day retention, and a custom-priced Enterprise plan with SSO, RBAC and VPC hosting. Hosted cost therefore scales with logged requests and retention, while self-hosted cost scales with infrastructure and staff time. Before comparing, confirm current pricing directly with each vendor, and model your log volume for a typical month and for peak load. Note that the Portkey page also lists a free self-hosted open-source option with unlimited logs, which changes the comparison.
Cost components for build, partner and SaaS AI gateway options
This table compares the cost components of running an AI gateway in-house, through an implementation partner, or on a hosted SaaS plan, and shows who pays for each one and whether it is one-time or recurring. Read each row as a question to put to your own quote: the software line is often free, while the infrastructure, identity integration and operations lines carry most of the ongoing effort. LiteLLM is MIT licensed outside its enterprise directory, but SSO beyond the free limit, audit logs and SCIM require an Enterprise license.
| Cost component | In-house | Implementation partner | Hosted SaaS |
|---|---|---|---|
| Software licensing | Open-source core is free to use. LiteLLM code outside its enterprise directory is MIT licensed. Enterprise features need a vendor license, paid directly by you. | Partner fee covers scoped setup and configuration. The software license is still paid by you to the vendor where an enterprise feature is needed. | Subscription paid to the SaaS vendor, usually recurring and often tiered by logged requests or retention period. |
| Infrastructure (PostgreSQL, Redis, compute) | Recurring cloud or data center cost for compute, PostgreSQL and Redis, paid by you. Sizing depends on request volume and instance count. | Usually the same recurring infrastructure cost, paid by you to your cloud or hosting provider. The partner may size and document it as part of scope. | Included in the subscription. Ask where the data plane runs and whether a private cloud or VPC option carries an extra recurring charge. |
| SSO and audit log licensing | Enterprise license for SSO beyond the free user limit, audit logs and SCIM, paid to the vendor on a recurring basis if you use those features. | Same vendor license, passed through or invoiced separately. Partner fee for configuring SSO and audit output is one-time or recurring depending on scope. | Often a higher tier. Confirm which SSO, audit log and RBAC features are included in the plan you quote. |
| Integration with identity and secrets | Internal staff time, one-time for the initial build of identity provider, secret manager and key rotation. Ongoing cost for maintenance. | One-time fixed-scope work to integrate identity and secret managers, with the client paying for the agreed deliverables. | Usually limited to the connectors the vendor supports. Custom integration with your identity provider or vault may be quoted separately. |
| Guardrail and provider configuration | Internal engineering time to define routing, model allowlists, budgets and guardrail rules. Recurring as providers and policies change. | One-time configuration work plus optional recurring review. Scope should list each provider and guardrail rule. | Configuration through the vendor console, often included. Custom guardrails or provider onboarding may be charged as add-on work. |
| Operations and upgrades | Recurring staff time for monitoring, patching and release tracking. Expect the team to own on-call and incident response. | Optional recurring retainer for upgrades, monitoring and testing. Confirm response times and what is excluded in writing. | Covered by the vendor's operations, usually recurring within the subscription. Confirm uptime commitments and maintenance windows in the contract. |
| Support | Internal support cost, plus any paid vendor support plan you choose to buy. | Support is included or sold as a retainer in the written scope. Ask what is covered after go-live. | Tiered vendor support, usually recurring. Confirm response targets in the contract rather than assuming them. |
What a self-hosted LiteLLM or APISIX stack costs to run
The main infrastructure cost of a self-hosted LiteLLM stack is the database and cache around the proxy, not the proxy itself. LiteLLM's production guide names PostgreSQL as the primary supported database, and virtual keys need PostgreSQL to store per-key, per-user and per-team budgets, TPM and RPM limits and model restrictions. Once more than one proxy instance runs, the guide requires Redis 7.0 or newer for shared rate limits, router state and cache. The guide suggests 1 vCPU and 4 GiB memory per worker and recommends one Uvicorn worker per pod on Kubernetes, scaling horizontally instead. Spend updates are batched to the database every 60 seconds at scale, so finance reports will lag live traffic by up to that interval. Each of these components is a server or managed service your team runs, patches, backs up and monitors, so the recurring cost is their operations as well as their hosting.
APISIX takes a different footprint. In traditional mode, the data and control planes share an etcd cluster, so you operate etcd alongside the gateway. In decoupled mode, separate data planes share that same etcd cluster. In standalone mode, APISIX needs no etcd and reads its configuration from conf/apisix.yaml or conf/apisix.json, reloading changes within a second. That removes a stateful cluster from the design, but it shifts configuration management into files or the Admin API, which your change process must version, review and back up. Choosing APISIX also means your AI-specific controls, such as budgets and virtual keys, may need to be built or added as plugins rather than configured out of the box. Confirm with the vendor or your integrator which of these controls your chosen version supports before you size the stack.
Sizing must come from your own traffic, not from the reference figures above. Ask your implementation partner to model peak concurrent requests, average prompt and response size, the number of teams and keys, log retention length and the number of model providers, then test a representative load against the proposed database and cache tiers. Confirm the high availability design for PostgreSQL and Redis, since a single database outage stops every application that depends on the gateway. Request the sizing assumptions in writing with the quote, so that cost changes can be traced to specific drivers such as log volume or replica count.
What moves the cost up or down
Most of the cost of an AI gateway comes from a handful of decisions made before purchase: how many requests and log days you keep, how many people need enterprise identity features, what audit evidence you must produce, how many providers you route to, and how much availability you need. Each card below shows which way a driver pushes the total and who pays for it under each option.
Logged request volume and retention
Every logged request adds storage, database writes and log review work, and retention multiplies that footprint. With self-hosted LiteLLM, the infrastructure and PostgreSQL storage are paid by your organization, so cost grows with traffic and days kept. With hosted gateway plans, the price is usually tied to recorded logs and retention window, so a higher log volume moves you to a larger tier. Ask any vendor for its pricing unit, its overage terms and how log retention is priced, and model your own request volume against each option.
Users who need SSO beyond the free limit
In LiteLLM, SSO for the admin UI is free for up to 5 users from v1.76.0, and an Enterprise license is required beyond that. The license cost is a recurring subscription paid to the vendor, and it scales with the headcount of administrators who sign in through SSO, not with request volume. Confirm the current license terms and whether the count applies to admin users or to all gateway users before you size the budget. Hosted plans often bundle SSO into a custom enterprise tier, so compare the quoted price against the same headcount.
Audit log and SCIM requirements
Audit logs, SCIM provisioning and organization and team RBAC sit in LiteLLM's Enterprise tier, so a regulated buyer who needs them pays a license fee on top of infrastructure. The cost is paid to the vendor for the license and to your team or partner for integrating SCIM with your identity provider and retaining audit records. If your evidence requirements are light, the cost stays lower. Ask which features are included in the quote, whether audit logs are retained by you or by the vendor, and who maintains the SCIM connector after go-live.
Number of providers and self-hosted models
Each model provider you route to adds a configuration to maintain, credentials to rotate, a pricing table to track and a failover path to test. Routing to self-hosted models on vLLM or Ollama moves inference cost onto your own GPUs and servers, which is a capital and operations cost rather than a per-token fee. The gateway software cost stays roughly flat as you add providers, but integration and testing effort grows with each one. Ask a partner to price the provider list as a fixed scope item, including the failover tests.
High availability and gateway instances
Running more than one proxy instance requires Redis 7.0 or newer for shared rate limits, router state and cache, and PostgreSQL remains the primary database. Each added instance consumes compute, and the production guide suggests 1 vCPU and 4 GiB memory per worker as a starting point to validate under your own load. Redundant PostgreSQL and Redis, load balancing and on-call coverage recur every month. Your organization pays the infrastructure and operations staff, or a retainer partner does. Ask for the topology, the failover test plan and who is paged at 2 a.m.
Identity, secrets and SIEM integration depth
Connecting the gateway to your identity provider, secret manager and security monitoring is often the largest one-time cost in a self-hosted build. LiteLLM Enterprise supports secret manager integrations with AWS, Azure, Google, HashiCorp Vault and CyberArk, and log export must reach your SIEM. Each integration is paid as implementation time, usually fixed scope, plus a recurring cost to keep it working through upgrades. Confirm which integrations are native, which need custom code, and whether the partner will test them after each gateway release.
How hosted AI gateway plans price by logged requests
Hosted AI gateway vendors typically meter their plans by recorded logs and how long those logs are kept. Using Portkey's published pricing as an example, the free Developer plan includes 10k recorded logs a month with 3-day log retention, and the Production plan is listed at $49 a month with 100k recorded logs and 30-day retention. The Enterprise plan is custom-priced and adds SSO, RBAC, private cloud and VPC hosting. Portkey also lists a free, self-hosted open-source option with unlimited recorded logs, which changes the cost question from subscription fees to infrastructure you run.
- Metered by logs and retention: Plan limits rise with recorded logs and the number of days logs are kept, so a team that needs longer audit retention moves up a tier even when request volume is unchanged.
- Enterprise features behind a custom quote: SSO, role-based access control and private or VPC hosting sit in the custom-priced tier, so the cost of identity and data residency is only known after a sales conversation.
- Vendor-run operations: The vendor operates the gateway, which removes patching and scaling work from your team but also places request and response data in the vendor's environment.
- Self-hosted option on the same vendor: The free self-hosted edition shifts cost to your own servers, databases and staff, with no per-log charge.
- Growth changes the bill: Log volume, retention and enterprise features all grow with usage, so model the cost at expected peak volume and at your required retention period.
Before you commit, confirm current prices, the log retention periods in each plan, where logs and prompts are stored, and the data residency terms for your region directly with the vendor.
A self-built gateway runs open-source LiteLLM or Apache APISIX in your own cloud account or data center, typically with PostgreSQL for virtual keys and spend records and Redis once more than one proxy instance runs. The software licence is free, but the cost moves to infrastructure, engineering time and operations. LiteLLM's production guide names PostgreSQL as its primary supported database, requires Redis 7.0 or newer for shared rate limits, router state and cache when several instances run, and suggests 1 vCPU and 4 GiB memory per worker.
- Infrastructure you pay for: Compute, managed or self-run PostgreSQL and Redis, load balancers, backups and log storage recur every month and grow with request volume and retention.
- Engineering time for setup: Routing, virtual keys, budgets, SSO and logging configuration, plus integration with applications, are one-time work whose effort depends on how many providers and teams you onboard.
- Enterprise licence for some features: LiteLLM's SSO is free for up to 5 users from v1.76.0, and beyond that an Enterprise license is needed. SCIM, audit logs, organization and team RBAC and secret manager integrations sit in the enterprise tier, so confirm which features your requirements need.
- Ongoing operations: Upgrades, security patching, incident response and retention cleanup recur and need named owners on your platform team.
- APISIX as a separate path: APISIX is Apache-2.0 licensed and can add AI gateway functions to an existing API gateway, which may reduce new tooling if your team already runs it.
Ask your team to estimate infrastructure, licence and staff hours in writing for each option, and confirm which LiteLLM features require an Enterprise licence for your use case.
In this model the gateway runs in your infrastructure, but an implementation partner handles part of the planning, configuration, integration, hardening and support. Your organisation keeps data control and pays the infrastructure and licence costs, while the partner's fees cover scoped work. DeployOpen plans, implements, integrates, hardens, backs up, upgrades, documents and supports self-hosted open-source platforms in the client's own infrastructure, and most engagements are fixed-scope with deliverables and timelines agreed in writing before work starts.
- Fixed-scope implementation fee: Architecture, deployment, identity integration and onboarding of initial teams are usually quoted from a written scope, so define the providers, teams and features covered before requesting a quote.
- Retainer for ongoing support: Upgrades, monitoring and continuous testing can be contracted separately, which converts recurring operations work into a predictable service cost.
- Infrastructure and licences stay with you: Compute, databases and any LiteLLM Enterprise licence are your costs and should be listed separately in any quote.
- Documentation and handover: Ask whether the scope includes runbooks and knowledge transfer so that your staff can operate the gateway after the engagement.
- Compliance evidence preparation: Partners can prepare configuration records and logs for assessments, but independent auditors issue any reports or certifications.
Request a written scope, a line-item quote separating partner fees from infrastructure and licence costs, and a statement of who holds administrative access after handover.
One-time setup versus recurring operating cost
The one-time work in a self-hosted AI gateway is design and integration effort: the architecture, the key and team model, identity integration, guardrail design, load and failover testing, and the documentation and evidence design that auditors will ask for. The key model deserves the most attention early. Decide how LITELLM_MASTER_KEY is stored, how virtual keys map to teams and applications, and which budgets and TPM or RPM limits apply to each. Identity integration is a separate line item if you need SSO for the admin UI beyond the free allowance, since LiteLLM's SSO is free for up to 5 users from v1.76.0 and an Enterprise license is needed beyond that. Confirm the current license terms with LiteLLM before you price this work, because the enterprise directory is licensed separately from the MIT-licensed remainder of the code.
The recurring work is where most of the operating cost sits. It includes running PostgreSQL for keys and spend data, running Redis 7.0 or newer once more than one proxy instance runs, patching both, and keeping backups and restore tests current. It also includes upgrades. LiteLLM supports only its four most recent stable minor versions, so you must plan a release cadence that stays inside that window, including regression tests against your own routing rules. Add on-call coverage for gateway outages, provider API changes that break routing, and regular budget reviews with finance. Each of these is a staffing decision as much as a technical one, and each should appear as a named role with hours in your own estimate rather than as a line that disappears into general platform work.
A hosted plan turns most of this recurring work into a subscription. Portkey's public pricing lists a Production plan at $49 a month with 100k recorded logs and 30-day log retention, and an Enterprise plan with SSO, RBAC, private cloud and VPC hosting priced on request. Those figures describe the vendor's subscription only, so compare both sides over the same period, for example 36 months, and include the internal staff time, cloud costs, and incident exposure on the self-hosted side. Also model log volume, since hosted plans are priced by recorded logs and your traffic growth changes the bill. Ask any vendor, in writing, where prompt logs are stored, for how long, and whether the logs leave your jurisdiction.
How to build your own AI gateway cost estimate
This sequence turns traffic, provider and compliance assumptions into a monthly cost model for a self-hosted gateway, so you can compare it fairly with hosted plans. The reader supplies the usage figures, and the output is a worksheet finance and engineering both accept.
Count applications and teams that call models
List every application, internal tool, script and team that will send requests to a model, including vendor features that call APIs on your behalf. Assign each one an owner, a cost center and an expected environment (development, test or production). The step is done when every caller maps to one team and one owner, and no caller is left unassigned. Keep this list, because virtual keys, budgets and per-tenant cost attribution all depend on it.
Estimate monthly request and log volumes
Pull current usage from existing provider invoices, application logs or a short pilot, and convert it into requests per month and average prompt and response size per caller. Record peak hours separately, since rate limits and worker sizing follow peaks rather than averages. Ask each team to confirm its numbers, because estimates made without measured data tend to drift. The output is a volume table with one row per caller, which you will reuse in the hosted plan comparison.
List required providers and self-hosted endpoints
Name every hosted model provider the gateway must reach, and every self-hosted endpoint (for example, a vLLM or Ollama service) it must route to. Note which models each caller may use, since model allowlists change both routing configuration and governance work. Confirm provider terms and the supported model list directly with each vendor and with the gateway documentation, because provider coverage changes. The result is a provider inventory that drives both integration effort and the number of routing rules to maintain.
Decide log retention and storage
Decide what the gateway records (metadata only, or prompts and responses), how long each record is kept and where it is stored. Logs are the largest source of storage growth and the basis of audit evidence, so this decision changes database sizing directly. Where the system may fall under the EU AI Act, confirm the retention period in the AI Act text with counsel, since Article 19(1) sets a minimum of six months unless other law provides otherwise. The output is a written retention schedule that engineering can size against.
List identity and audit requirements that trigger enterprise licensing
Document which administrators need single sign-on, whether group sync through SCIM is required, which teams need role-based access, and whether you must produce audit logs or restrict admin access by IP address. In the LiteLLM documentation, SSO is free for up to 5 users from v1.76.0, and SSO beyond that, SCIM, audit logs, RBAC and IP allowlisting sit in the Enterprise license. Get written license terms and the current feature matrix from the vendor before you build the model. The output is a list of required features mapped to a license line item or to a free alternative.
Size PostgreSQL, Redis and compute
Use the LiteLLM production guide as a starting point: PostgreSQL as the primary database, Redis 7.0 or newer once more than one proxy instance runs, and a baseline of 1 vCPU and 4 GiB memory per worker. Scale instance count from the peak volumes in step two, add headroom for failover and plan for one Uvicorn worker per pod on Kubernetes. Validate the sizing with a load test in a non-production environment before you commit to a number. The result is a bill of materials for cloud or on-premises infrastructure, with recurring costs listed separately from one-time setup.
Price the operations model in staff hours
Estimate the staff hours needed for installation, integration with identity and monitoring, upgrades, backups, key rotation, incident response and on-call coverage. Assign each activity to an internal role or to an implementation partner under a written scope, and note who is paid for what. Split costs into one-time (design, integration, initial hardening) and recurring (upgrades, support, infrastructure, licenses). The output is a staffing table with hours per month, which you can convert to cost using your own internal rates.
Compare hosted plan limits against the same volumes
Take the volume table from step two and the log retention schedule from step four, and check them against each hosted gateway plan's limits, such as recorded logs per month, log retention days and the plan that includes SSO, RBAC or private hosting. Portkey's public pricing lists a Developer plan with 10k recorded logs a month and 3-day retention, a Production plan at $49 a month with 100k recorded logs and 30-day retention, and a custom Enterprise plan. Ask each vendor for a written quote covering your actual volume, retention and data residency needs, then place the hosted quote beside the self-hosted model.
Why log retention changes the cost comparison
Retention and log volume decide a large part of the cost comparison, because every model call that the gateway records is data someone must store, search and keep. Hosted gateway plans meter by recorded logs and retention window, so a plan that looks cheap at low volume can rise sharply once every request is kept for the period your obligations require. Portkey's public pricing, for example, lists a free Developer plan with 10k recorded logs a month and 3-day retention, and a Production plan at $49 a month with 100k recorded logs and 30-day retention (per the Portkey pricing page). Compare those limits with your expected request count and the retention period you actually need, not with the lowest tier.
EU AI Act Article 12(1) requires high-risk AI systems to technically allow automatic recording of events over the system's lifetime, and Article 19(1) requires providers to keep those logs for at least six months unless other law provides otherwise. Whether your systems are in scope, and how long each log category must be kept, are questions to put to counsel before you size anything. The answers can differ by system, by role (provider or deployer) and by sector, and they may require longer retention than six months in some cases.
With a self-hosted gateway, the cost moves to your own infrastructure. LiteLLM stores spend and request data in PostgreSQL, so you pay for database storage, backups, replication and the time to run retention jobs, and you own the decision to delete records on schedule. Ask your team to estimate log volume per model call, multiply by request count and retention period, and confirm storage and backup costs with your platform or cloud team. Those figures, not the software license, will often decide the build versus SaaS outcome.
AI gateway cost questions buyers ask before hiring help
How much does a self-hosted AI gateway cost to run?
The software license is often zero, but the total cost is made of components you pay for separately: the gateway software itself, PostgreSQL and Redis, compute for the gateway workers, log storage, a load balancer or ingress, identity integration, and the people who build and operate it. Open-source LiteLLM is MIT licensed outside its enterprise directory, so the software cost is the infrastructure plus labor. Enterprise features such as SSO beyond the free user limit, SCIM, audit logs and JWT authentication add a license cost, which is quoted by the vendor. Recurring costs include infrastructure, model provider usage, log retention, upgrades and on-call support. Ask any partner for a quote that lists each component, who pays for it, and which items are one-time and which recur.
How long does it take to set up an AI gateway?
Duration depends on scope, so there is no single answer that holds for every organization. The main drivers are how many model providers and self-hosted models must be connected, whether SSO and team mapping are required, how many environments you need, how many guardrail and logging rules must be defined, and whether your Kubernetes, PostgreSQL and Redis platforms already exist. A fixed-scope engagement sets the deliverables and timeline in writing before work starts, so ask for that document rather than a general estimate. Confirm in writing what is included in the first phase, what is tested before production, and what acceptance criteria the gateway must meet.
What does the gateway project need from our team?
Expect to supply several inputs before work begins. You need named identity owners for SSO and team mapping, a list of applications and teams that will call the gateway, and the model providers and self-hosted models that are approved. Traffic data helps size rate limits and budgets. Retention decisions are essential: you must decide how long prompts, responses and logs are kept, where they are stored, who can read them and how they are deleted. You also need infrastructure access in your own cloud account, data center or air-gapped network, plus a decision maker who signs off on guardrail rules. Missing decisions at the start are the most common reason scope grows later.
Can our own engineers build an AI gateway without a partner?
Yes, many teams can, provided they already have working skills in PostgreSQL administration, Redis, Kubernetes and secrets management, plus an owner for security review. LiteLLM's production guidance names PostgreSQL as the primary supported database and requires Redis 7.0 or newer once more than one proxy instance runs, so these are prerequisites rather than optional extras. Before building, settle the enterprise license question: SSO for more than five users, SCIM, audit logs and some secret manager integrations require an Enterprise license in LiteLLM. Decide whether APISIX fits better if you already run an API gateway. Internal teams should also plan for upgrades, since LiteLLM maintains support for the four most recent stable minor versions.
When is a hosted AI gateway SaaS cheaper than self-hosting?
A hosted plan is usually cheaper when volume is low, you have no platform team to run PostgreSQL and Redis, and your data rules allow external processing of prompts and logs. Hosted plans are typically priced by logged requests and retention period, so cost rises with traffic and with how long logs are kept. For example, Portkey's pricing page lists a free Developer plan with 10k recorded logs a month and 3-day retention, and a Production plan at $49 a month with 100k recorded logs and 30-day retention, while its Enterprise plan is custom-priced. Self-hosting becomes more attractive as volume, retention needs or residency requirements grow. Compare total cost over your expected log volume and retention, not the entry price.
Which LiteLLM features need an Enterprise license?
LiteLLM is MIT licensed except for content under its enterprise directory, which uses the license in that folder. Per LiteLLM's documentation, SSO for the admin UI is free for up to 5 users from version 1.76.0, and an Enterprise license is needed beyond that. Enterprise adds SCIM, JWT authentication with identity provider tokens, audit logs, organization and team RBAC, IP allowlisting, secret manager integrations for AWS, Azure, Google, HashiCorp Vault and CyberArk, and team-level guardrails and logging. Confirm current terms with the vendor before you design around any feature, because license scope affects both cost and the compliance evidence you can produce.
What does an implementation partner typically cover in the fee?
A fixed-scope engagement usually covers planning, deploying the gateway in your own infrastructure, configuring providers, virtual keys, budgets, routing and fallback, integrating identity, setting up guardrails and logging, hardening, backup, documentation and handover. DeployOpen plans, implements, integrates, hardens, backs up, upgrades, documents and supports self-hosted open-source platforms inside the client's own environment. Ongoing support, upgrades and continuous testing are available as retainers. Confirm in writing which items are in the fixed scope, which need a separate change request, and whether enterprise licenses are purchased by you or included. DeployOpen does not publish a price list, so a quote follows from the written scope.
How DeployOpen works on this
DeployOpen is a Chainscore Labs company that plans, implements, integrates, hardens, documents and supports self-hosted open-source platforms inside the client's own infrastructure. For AI gateway cost, we start from a written scope that compares build, partner and hosted options on the same workload, so the quote reflects the platform, the integrations and the operating model you actually need.
How we scope
- Inventory model providers, teams, applications and expected request volumes that the gateway must serve.
- Define the written scope for the self-hosted stack, including PostgreSQL, Redis, identity integration and log storage.
- Identify which features need an enterprise license, such as SSO beyond the free user limit, SCIM and audit logs.
How we build
- Deploy the chosen gateway, such as LiteLLM or Apache APISIX, in your cloud accounts, data centers or air-gapped network.
- Configure virtual keys, team budgets, rate limits, routing, fallback and guardrails according to the agreed scope.
- Test failover, budget enforcement and log capture, and document the results against the agreed acceptance criteria.
How we hand over
- Deliver architecture diagrams, configuration records, runbooks and a cost model that your finance team can review.
- Document which cost components are one-time, such as implementation, and which recur, such as licenses, hosting and support.
- Transfer administration knowledge to your operators so the gateway can be run without the implementation team.
How we support
- Offer ongoing support, upgrades and continuous testing as a retainer after handover, scoped separately in writing.
- Review gateway version changes, license terms and log retention settings with your compliance and platform leads.
- Re-price the operating model when request volumes, provider mix or enterprise feature needs change.
