Compare·Enterprise AI gateway
Continuum
Multi-agent workbench
VS
Portkey
Enterprise AI gateway

Continuum vs Portkey

Both put policy in front of AI spend. One sits on the network path; the other sits inside the tool developers open.

Portkey is an enterprise AI gateway. One OpenAI-compatible endpoint fronts what its site calls 1,600+ models, and around that endpoint it adds observability, prompt management, 40+ pre-built guardrails, semantic caching, role-based access control, budgets, and SSO. The open-source gateway is MIT licensed with roughly 12.8k GitHub stars and a 122kb footprint. In 2026 the company changed shape: Palo Alto Networks announced its acquisition on 30 April 2026 and completed it on 29 May 2026, making Portkey the foundational AI Gateway inside Prisma AIRS. Continuum is not a proxy. It is the free workbench where developers actually run Claude Code, Codex, Cursor, Grok, and OpenCode under their own subscriptions, with per-session worktree isolation, plan gates, per-hunk diff review, and PR merge in the session. Its governance layer attaches there: org model policy, weekly spend caps, and approvals over the surface where coding agents run. The two answer different questions. Portkey asks what your applications may send to a model. Continuum asks what your engineers' agents may do to a repository, and what that cost.

Download for Mac All comparisons
Workflow Monday path Matrix Decide Money FAQ

Updated 2026-08-03 · Mac stable · Win/Linux desktop beta

Choose Continuum when

Pick Continuum when the AI surface you need to govern is coding agents on developer machines, and you want model policy, weekly spend caps, approvals, and per-repo cost without standing up or routing through proxy infrastructure.

Choose Portkey when

Pick Portkey when a platform team must govern LLM traffic from many internal applications and services, and needs one endpoint with guardrails, prompt versioning, caching, log retention, and per-key budgets across every model your company calls.

Snapshot direct alternative
Dimension Continuum Portkey
Control point The workbench agents run in The network path model calls take
Primary buyer Engineering org running coding agents Platform team governing LLM traffic
Model reach Claude · Codex · Cursor · Grok · OpenCode 1,600+ models behind one endpoint
Policy shape Org model policy · weekly caps · approvals Guardrails · budgets · RBAC · SSO
Repo contract Worktree + branch · plan gate · diff · PR Not in scope for a gateway
Cost view By repo, provider, model, and day By key, model, and request with log retention
Ownership Free app · optional hosted inference MIT gateway · $49/mo Production · now Palo Alto Networks
Mac stable · web · iPhone · Watch · Win/Linux desktop beta · free app
01

Govern the network path vs govern the workbench

In Continuum

Give an engineering org control over agent work

09:20

A member starts a Codex session on a ticket. Org model policy allows the model; the session opens in its own worktree and branch.

11:05

Plan mode finishes read-only. A reviewer approves it from the phone, and the same session moves into the write-capable run.

15:40

The member hits the weekly spend cap mid-afternoon and files an approval request instead of quietly burning into next week's budget.

17:30

The lead reads spend by repo, provider, model, and day in Usage analytics and sees which repository consumed the window.

In Portkey

Put one governed endpoint in front of company LLM traffic

09:20

Point an internal service at the Portkey gateway and get retries, timeouts, fallbacks, and load balancing without changing the calling code.

11:05

Turn on PII redaction and the guardrails the compliance team asked for, so sensitive fields are stripped before a request leaves.

15:40

Move the team's prompts into managed templates with versioning and variables, then compare versions in the playground.

17:30

Read the observability dashboard for logs, traces, and alerts, and set per-key budgets and rate limits for the next quarter.

02

Monday path

Monday with a gateway and a workbench side by side

Install, connect, first session - steps you can run the same day.

01

Write down which AI traffic you are actually trying to govern. Application and service calls to models is Portkey's job. Coding agents running on laptops is Continuum's.

02

Check how your engineers authenticate today. A Claude or ChatGPT subscription logged in through the CLI never crosses a company proxy, so a gateway cannot see, cap, or log any of it.

03

Stand up the Portkey open-source gateway or the Developer tier for free, point one internal service at it, and confirm the guardrails, cache, and log retention you need are on the tier you plan to buy.

04

Note the log ceilings while you are there. The Developer tier records 10,000 logs per month with 3-day retention; Production records 100,000 with 30-day retention.

05

Install Continuum on one team's Macs, start a worktree session per ticket, and watch the plan gate hold the agent read-only until someone approves.

06

Turn on org model policy and a weekly spend cap for that team, then send one approval request through so you have seen the whole loop.

07

At the end of the week, read Portkey's dashboard for application traffic and Continuum's per-repo, per-provider, per-model, per-day ledger for coding spend. Do not try to add them into one number; they meter different things.

08

Decide the split deliberately. Most orgs keep the gateway for product inference and keep the workbench for engineering, rather than forcing one tool to cover both.

How Continuum enforces policy without a proxy

How Continuum governs agents that never touch a proxy

  • Continuum launches the provider's real CLI under the developer's own login, so policy has to live at the workbench rather than on a network hop the CLI never takes.
  • Org model policy decides which providers and models a member may start a session with, enforced at spawn time instead of at request time.
  • Weekly spend caps and approval requests sit on the same object, so a member who hits the ceiling asks for headroom rather than silently failing a turn.
  • Every managed session gets its own git worktree and branch, which is the containment boundary a gateway cannot provide at all.
  • Plan mode stays read-only until an explicit approval, and the approval can arrive from iPhone, web, or Mac against the same live session.
  • Local agent logs are priced into tokens and dollars by repo, provider, model, and day without an inference proxy in the path for BYOK sessions.
  • When you do want gateway-shaped billing, Continuum's hosted inference exposes an OpenAI-compatible endpoint at https://continuumcode.ai/v1 and an Anthropic-compatible bare origin with cont_sk_ keys.
03

Gateway and workbench control matrix

Cells use product nouns on both sides. Continuum is the multi-provider workbench; they win where their loop is the product.

Capability Continuum Portkey
Governs coding agents on developer machinesSubscription CLI logins never cross a proxy Org model policy at session spawn Only traffic routed through the gateway
One endpoint across many model providersBreadth of the catalog is Portkey's category win Hosted inference is OpenAI and Anthropic compatible Site states 1,600+ models behind one API
Content guardrails on requests and responses No PII or toxicity filtering layer 40+ pre-built guardrails including PII redaction
Prompt management and versioning3 templates on Developer, unlimited on Production Not a prompt template product Templates, versioning, variables, playground
Request observability with log retention10k logs at 3 days free; 100k at 30 days on Production Local cost and token history, not request logs Logs, traces, feedback, filters, alerts
Semantic and simple caching No response cache layer Simple caching free; semantic caching on Production
Per-session git worktree and branch isolationThe containment boundary for autonomous edits Real worktree plus branch per managed session A gateway has no repository layer
Plan approval before an agent writes Read-only Plan mode with explicit approve and run Not a feature of a request proxy
Per-hunk diff review and PR merge in session Stage, revert, create PR, watch checks, merge Out of scope by design
Spend caps and budget enforcementBoth enforce, at different layers Weekly caps per member with approval requests Granular budget and rate limits on Enterprise
Live provider quota gauges 5h and weekly subscription windows read live Meters gateway usage, not a provider subscription window
Self-host the control plane Agents run on your own enrolled hosts MIT gateway self-hosted; private cloud and VPC on Enterprise
Compliance certifications published Not asserted on this page SOC2 Type 2, GDPR, HIPAA, custom BAAs on Enterprise
04

Where the two control planes separate

01 · Interception point

A proxy can only govern the traffic that goes through it.

This is the load-bearing difference and it is not a criticism of Portkey. A developer running Claude Code under a personal Claude subscription authenticates to Anthropic directly. That traffic never takes a company network hop, so a gateway records none of it, caps none of it, and cannot apply a guardrail to it. Continuum governs at the point the agent is launched, which is why its model policy and weekly caps reach the sessions a gateway is structurally blind to.

02 · What the policy is about

Portkey polices payloads. Continuum polices repository actions.

Portkey's guardrails ask whether a request contains PII, whether a response is safe, and whether the caller is within budget. Continuum's gates ask whether an agent may write to this repository at all, on which branch, after whose approval, and with which hunks accepted. Those questions do not substitute for each other, and a team with both problems genuinely has two problems.

03 · Cost accounting

Per-key metering and per-repo attribution answer different questions.

Portkey attributes spend to keys, models, and requests, with log retention as the tiering axis: 10,000 recorded logs a month on Developer, 100,000 on the $49 Production tier, more on Enterprise. Continuum attributes local agent history to repo, provider, model, and day, so the answer to why the weekly window collapsed is a repository name rather than a key id. See reducing Claude Code costs for how that reads in practice.

04 · Company shape

Portkey is now part of Palo Alto Networks. Weigh that either way.

Palo Alto Networks announced the acquisition on 30 April 2026 and completed it on 29 May 2026, positioning Portkey as the foundational AI Gateway inside Prisma AIRS. For a security-led enterprise buyer that is a strong signal: the gateway is now backed by a large security vendor and sits inside a broader AI runtime security story. For a small team it changes the sales motion and the roadmap owner. Either way, buy against the pricing and terms you can read today, and note that the MIT-licensed gateway remains self-hostable.

05 · What ships free

Both have a real free tier; they are free at different things.

Portkey's open-source gateway is MIT with no request limit when you host it, and the managed Developer tier is free with 10,000 recorded logs a month. Continuum's workbench is free outright on Mac, web, iPhone, and Watch, running under the subscriptions you already pay for, with hosted inference as a separate optional plan. Neither free tier is a trial; both are the actual product with ceilings on a specific axis.

05

Coexistence

Stack recipe

How people run both.

They can coexist cleanly, because they intercept at different points. Route your product's inference through Portkey and keep its guardrails, prompt templates, and log retention. Run your engineers' coding agents through Continuum, where the branch boundary, the plan gate, and the diff review live. If you want coding-agent traffic on a gateway too, Continuum's own hosted inference speaks the same OpenAI-compatible and Anthropic-compatible shapes, so it can be the endpoint a client points at rather than a second hop in front of one.

Download Continuum Providers
06

Pick by what you are governing

Continuum

Choose Continuum when the surface is coding agents

  • The AI surface you must govern is coding agents on engineers' machines, not application traffic.
  • Your developers authenticate with their own Claude or ChatGPT subscriptions, so no proxy sees that spend.
  • You need a branch boundary, a plan gate, and diff review, which no gateway provides.
  • You want cost attributed to a repository rather than to an API key.
  • You want governance running this week without deploying or routing through new infrastructure.
  • You would rather add a flat-fee hosted endpoint later than operate a proxy now.
Portkey

Choose Portkey when the surface is application LLM traffic

  • Many internal applications and services call models and need one governed endpoint.
  • Compliance requires content guardrails such as PII redaction on requests and responses.
  • Prompt templates need central versioning, variables, and a playground across teams.
  • You need request-level logs, traces, and alerts with a defined retention window.
  • Caching, fallbacks, retries, and load balancing across providers are a reliability requirement.
  • You want SOC2 Type 2, GDPR, HIPAA, BAAs, and a private cloud or VPC deployment.
07

What you pay and what meters

Continuum

App + your labs

$0 for the workbench on Mac, web, iPhone, and Watch, running under the Claude, ChatGPT, Cursor, or Grok subscriptions you already pay for, with no cut taken on those sessions. Optional hosted inference is Plus at $25, Max 100 at $100, Max 200 at $200, and Ultra at $500 per month, carrying weekly allowances of $25, $100, $200, and $1,000.

Portkey

Their bill

Portkey's open-source gateway is MIT licensed and free with no request limit when you host it. The managed Developer tier is free forever with 10,000 recorded logs a month, 3-day log retention, and 3 prompt templates. Production is $49 a month for 100,000 recorded logs, then $9 per additional 100,000 up to 3 million, with 30-day log retention and 90-day metrics. Enterprise is custom, covers 10+ million logs a month, and adds SSO, granular budget and rate limits, private cloud or VPC hosting, and SOC2 Type 2, GDPR, and HIPAA compliance.

How to compare

Total cost of work

The meters are different, so the comparison is not per-dollar. Portkey prices recorded observability on top of traffic you route through it. Continuum charges nothing for BYOK sessions and prices only optional hosted inference, as a flat monthly fee with a weekly allowance rather than per request.

Portkey plans checked 3 August 2026 against their pricing page. Vendors reprice often - if you spot a stale figure, tell us and we will fix it. Continuum never marks up a session it runs under your own login.

Continuum ladder: Free app · Plus $25/mo · Max 100 · Max 200 · Ultra - full pricing. Also analytics, multi-account, devices.

08

Questions

Continuum vs Portkey.

Deep dives: docs, providers, sessions.

Portkey is an enterprise AI gateway. It puts one OpenAI-compatible endpoint in front of what its site describes as 1,600+ models and adds observability, prompt management, guardrails, caching, role-based access control, budgets, and SSO. Its gateway is also available as MIT-licensed open source with around 12.8k GitHub stars.

Yes. Palo Alto Networks announced the acquisition on 30 April 2026 and completed it on 29 May 2026. Portkey now serves as the foundational AI Gateway inside Prisma AIRS. The open-source gateway remains MIT licensed and self-hostable.

The self-hosted open-source gateway is free with no request limit. The managed Developer tier is free with 10,000 recorded logs a month and 3-day retention. Production is $49 a month for 100,000 recorded logs, with $9 per additional 100,000 up to 3 million and 30-day log retention. Enterprise is custom priced for 10+ million logs a month.

Only when the thing you need to govern is coding agents. Continuum does not provide guardrails, prompt templates, semantic caching, or request log retention. It provides org model policy, weekly spend caps, approvals, worktree isolation, plan gates, diff review, and per-repo cost for the agents developers run. If you need to govern application LLM traffic, that is a gateway's job.

Not when developers are signed in with their own Claude subscriptions. That traffic goes straight to the provider and never crosses your proxy, so it appears in no gateway dashboard. Continuum reads local agent history instead and prices it by repo, provider, model, and day, which is why it can report on spend a gateway cannot observe.

That is a gateway-to-gateway comparison and both are legitimate answers to the routing question. Portkey's published differentiators are its managed observability tiers, prompt management, 40+ pre-built guardrails, and the enterprise compliance and deployment options. Evaluate them against your log retention needs and self-hosting appetite. Neither choice affects whether you also need a workbench for coding agents, which is the axis Continuum is on.

Yes, and it is the common shape. Keep Portkey in front of your product's inference, and run engineering's coding agents in Continuum. If you also want coding-agent traffic metered on a gateway, Continuum's hosted inference exposes an OpenAI-compatible endpoint and an Anthropic-compatible bare origin with cont_sk_ keys, so a client can point at it directly.

Begin

Run your agents
in Continuum.

Free app. Your subscriptions. Optional hosted inference. Mac stable - Windows and Linux desktop are beta.

Download for Mac See pricing

vendor-neutral · local-first · multi-device