Blog
Field notes from the gateway.
Engineering decisions, model evaluations, and the occasional opinionated essay.
Two presets isn't a feature. It's a default.
ByteSpike ships with two curated model presets — Global and China — and every account opens to one of them already applied. We get asked why we didn't just let admins pick from 30 models on day one. Here's the rationale.
5 minDOSIA × ByteSpike: one agent, every capability
You shouldn't have to keep three apps open to get a marketing kit done. DOSIA's main brain now picks up image generation, image analysis, video, and 'use a different LLM' as native capabilities — one OAuth connect to ByteSpike, then you just type what you want.
7 minHow "failures don't bill" actually works
Every ByteSpike call carries a pre-flight credit reservation and a settle-on-success commit. Reservations expire if the upstream errors out, the user cancels, or the gateway times out — the account ledger only debits when there's an asset or a token stream that landed.
5 minMarketing, Console, Docs — why ByteSpike ships three apps from one repo
ByteSpike's website is three separate applications behind three subdomains: bytespike.ai is marketing, console.bytespike.ai is the customer portal, docs.bytespike.ai is Mintlify. They share design tokens, locale data, and deploy strategy — but their release cadence, attack surface, and bundle weight diverge enough that smashing them together would have cost us more than living with the seams.
5 minLive channel vs scaffold estimate — what the badge on /pricing means
Every row on the /pricing endpoint table carries a small status badge. "Live channel" means the price came from the gateway's admin export — what we actually charge today. "Scaffold estimate" means we listed a conservative public-list-price approximation until that endpoint's channel is wired up. The split is deliberate; here's why we surface it instead of hiding it.
4 minWhy ByteSpike doesn't ship a chat UI
Every multi-provider gateway eventually gets asked, "where's your chat playground?" — usually right before someone offers to build it. We've turned that question down five times so far. Here's the calculus.
4 minNext 15 SSG cache on a self-hosted nginx — what we learned the hard way
We spent a day staring at stale prod HTML before noticing that Next.js 15 emits `s-maxage=31536000` by default on every prerendered page, that `export const revalidate = N` doesn't change the emitted Cache-Control header (only the ISR background timer), and that our self-hosted nginx had no idea any of this had happened. Here's what fixed it.
4 minFive mistakes we made building a multi-provider gateway
We didn't get the shape of ByteSpike right on the first try. Here are five decisions we walked back after they hit production — each one cheap to write down, each one expensive to live through.
6 min5-Minute ByteSpike integration — from DOSIA download to first chat
Five steps, ~5 minutes: install DOSIA, sign up for ByteSpike, mint an API key, wire the provider into DOSIA Settings, and send your first prompt. By the end you'll be routing through one key to Claude / GPT / Gemini / DeepSeek and watching the cost land at sub-fen.
5 minDOSIA × ByteSpike — one team, two products, one workflow
DOSIA is the desktop client. ByteSpike is the multi-provider gateway. They ship separately on purpose — but the combination is the shape our team actually wanted: one account, one key, one bill, every frontier model.
5 minOne key for image, video, audio, and embeddings
ByteSpike's multimodal surface sits behind the same Anthropic-shape key as text — image, video, embeddings, rerank. Here are the three patterns we use ourselves and the failure-doesn't-bill semantics that make experimenting cheap.
6 minOne key, two protocols: why ByteSpike routes DeepSeek through both OpenAI and Anthropic shapes
DeepSeek's HTTP API quietly speaks both protocols. We wrapped both behind one ByteSpike key so your agent code and your chat code can use the same model without holding two credentials.
4 minDon't charge for failures: the engineering behind a one-sentence billing promise
"Failures don't bill" fits in a footer. Implementing it correctly across nine providers, two protocols, sync and streaming, sync and async, text and pixels — that is several thousand lines of code and one published refund policy.
7 minBuilding an AI customer support router with ByteSpike
How to chain Haiku for triage, Sonnet for response, and Opus for escalation behind one key with predictable costs and cache-read savings.
8 minThirteen typography decisions for a bilingual marketing site
We just shipped brand v2 across the ByteSpike marketing site — new wordmark, new gradient, new sparkle motif. The visible part took a week. The typography micro-work took longer than that, and you're supposed to never notice it. Here's what actually changed, and why.
7 minIntroducing ByteSpike: one key, all the world's models
ByteSpike is a multi-model gateway with one Anthropic-compatible key, native CJK pricing, and a transparent rate card. We built it because we got tired of writing nine different SDKs.
5 minByteSpike vs aireiter and b.ai: how the gateway choice changes
Three model gateways serve overlapping markets. The differences are concrete: how you authenticate, how billing works, which models are first-class, and what happens when calls fail. Here's what we'd say if we were the customer.
7 minFive workflows that ship faster on DOSIA + ByteSpike
DOSIA is the desktop AI companion we built on top of ByteSpike. Together they cover the loop from prompt design to production deployment. Five concrete workflows where the integration removes a step you'd otherwise build by hand.
8 minChoosing the right image and video model: a practical guide
Nine image models and eight video models behind one ByteSpike key. They aren't interchangeable — each has a brief it owns and a brief it loses on. Here's the call-it-by-name decision tree.
7 minByteSpike's roadmap: one key, all the world's models
Where ByteSpike is headed over the next four quarters. Four phases, each a concrete addition rather than a vague north star.
6 minWhy we built ByteSpike
Every team we know is gluing two or three model providers together with bash scripts. We got tired of it. So we built one gateway, one key, and one billing surface — and we made it Anthropic-compatible because that's the API we actually like writing.
6 minWhat we built when we stopped building chat
Every AI startup builds a chat UI in week two. We almost did. Then we counted how many chat UIs already exist and pivoted: ByteSpike for the API, DOSIA for the no-code path, nothing in the middle. Here's what fell out of that decision and why we'd make it again.
5 min23 endpoints, one key, 80 ms median overhead
ByteSpike fans out to 23 model endpoints across image, video, text, and async-job categories — through a single Anthropic-shape API. The interesting question isn't routing; it's the latency budget. Here's how we keep gateway median overhead under 80 ms while still doing OAuth pool stickiness, retry semantics, and per-request quota math.
6 minThe Anthropic-protocol bet
We picked the Anthropic Messages API as our default protocol. Here's why we think tool_use, cache_control, and thinking are the API surface most worth keeping pristine — and how we shim everything else without losing fidelity.
4 minDOSIA Monthly Brief · 2026-04 · ByteSpike integration + 5-layer file system goes live
April was the month DOSIA stopped being a single-account client and became a real multi-tenant agent host. ByteSpike provider, a five-layer file system from session to enterprise, and a native-binary SDK that drops startup by ~30%.
6 min