On October 15, DeepSeek's chat service and two of its APIs went down. The chat service was fully unavailable for a period, the DeepSeek V4.1 Flash API and the V4 Pro API were partially interrupted, and performance degraded on the website and the API at the same time. File upload and search were not affected.
One distinction matters before the timeline. ServBay AI Gateway is not a switcher, it is a gateway. A switcher answers which provider to use. A gateway answers how traffic moves between providers, and keeps that decision out of every tool's configuration. The October 15 incident is a practical way to see why the difference shows up in the middle of a failure.
What the status page recorded
At 08:11, the DeepSeek status page initially showed that the problem was being investigated. At 09:06, the page was updated to say the cause had been identified and a fix was being implemented. At 12:49, the fix went live and the service entered monitoring. At 13:04, the incident was marked resolved and service was restored.
| Time | Status page entry |
|---|---|
| 08:11 | Problem under investigation |
| 09:06 | Cause identified, fix being implemented |
| 12:49 | Fix live, service under monitoring |
| 13:04 | Incident resolved, service restored |
Nearly four hours of repair time
Nearly four more hours passed between identifying the cause and restoring service. Anyone writing copy, debugging code or drafting a proposal with DeepSeek was working around the same gap.
The apology and the remediation notice
After service was restored, DeepSeek published an apology and a compensation notice. The company said it has launched a comprehensive review and remediation effort focused on service stability, and will systematically comb through and eliminate single points of dependency to prevent a repeat.
A single point of dependency behaves like this. When one critical link fails, other services that rely on it can fail along with it. DeepSeek has not disclosed the specific trigger of this incident or which link failed, and a full technical explanation has not been published.
Compensation in credits
The compensation plan is the concrete part. DeepSeek refunds, in the form of credits, the full amount of each account's API consumption between October 8 and October 14. Credits arrive automatically and are applied first to subsequent API calls. An account that spent 100 yuan on API calls across those seven days receives 100 yuan in credits, with no application required. Developers who use the DeepSeek API can check the usage page on the DeepSeek open platform to see whether the compensation has arrived.
This is not the first incident of its kind
The status page history records 13 incidents since September 2, ranging from degraded performance to partial outages, with two occurring on October 12 alone. Availability figures from July through October are 99.74% for the V4 Pro API, 99.47% for the V4.1 Flash API, and 99.78% for the chat service.
Provider-side fixes and client-side exposure
Eliminating single points of dependency is a fix on the provider's side. The same idea applies one layer down. A project whose AI calls all point at a single channel inherits that channel's availability, and no amount of remediation upstream changes that.
Why manual incident response has a ceiling
Three responses are common when an outage hits. Switch to another API key by hand. Edit the base URL in code. Restart the service and try again.
All of them require a person watching. Switching a channel means changing configuration, and afterwards someone has to remember to change it back. With several projects and several coding tools running at once, the edit has to be repeated in every place. An outage that lands at night, or halfway through a task, is harder to contain.
A more workable arrangement puts channel switching behind one unified entry point, so tools and code connect to that entry point and nothing else.
ServBay AI Gateway as a gateway, not a switcher
ServBay runs as an AI development management tool, and the local AI gateway is the part of it that handles model traffic. Scattered model configuration is consolidated into one application and one local entry point.
One entry point for every model
Tools such as Claude Code and Codex only need their base URL pointed at the gateway. Switching models or providers afterwards happens inside ServBay and the tools themselves stay untouched. The gateway converts between the Anthropic, OpenAI and Gemini API formats, so models from any vendor can sit behind the same entry point, and local Ollama models are no exception.
Adding a provider means picking one, entering a key and saving it. There are currently native integrations with 12+ providers, among them OpenAI, Anthropic, Gemini, DeepSeek, Kimi, Zhipu GLM, Qwen, Baidu Qianfan and Tencent TokenHub, plus support for any provider compatible with the OpenAI API.
Each provider supports two levels of connectivity testing. Without a key, the test checks whether the endpoint is reachable. With a key, it checks whether the credentials are valid. Problems surface at configuration time, rather than during an outage when a backup channel turns out never to have been usable.
A setup like this fits developers who run multiple tools, multiple channels or multiple projects at the same time. Someone using a single tool and a single channel, where an occasional interruption does not block the work, does not need it yet.
Channel failover and where it stops
The same model can be wired to multiple channels, for example the official DeepSeek channel plus other platforms that also serve that model. Three routing modes are available. Fixed mode always uses the same channel. Failover mode uses the first channel first and, on failure, tries the backup channels in order. Balanced mode rotates requests across channels and is a VIP feature.
When the primary channel is rate-limited or down, the gateway switches to a healthy backup channel. There is a boundary to this mechanism. Failover only applies between channels that can serve the same model, and different models are never mixed. The gateway does not choose a model on the caller's behalf. When a different model is needed, the tool still specifies it.
Virtual keys issued per project
Projects and downstream tools call through local virtual keys, and the upstream real key is not exposed to them. A key can be issued per project or per tool, and each can be revoked individually. If one project's key leaks, revoking that key is enough and the real key is unaffected. Real keys are stored encrypted on the machine and never uploaded.
CLI tools and cost visibility
ServBay supports one-click takeover of Claude Code, Codex, opencode, Crush, Qwen Code, Kimi CLI and CodeBuddy. There is no need to edit export environment variables by hand or dig through .zshrc. One click in the interface sends these tools' requests through the local gateway, and usage and cost for each channel sit in one place, so reconciling several projects no longer means working through each provider's billing separately.
A setup that survives the next outage
Three steps to get running
- Add a provider. Each provider's real key is entered once and stored encrypted on the machine, never uploaded.
- Issue virtual keys per project or per tool.
- Point the tools' base URL at the gateway. After that, switching models or providers does not require touching the tools.
Two more steps aimed at failures
- Add at least two channels for the same model, one of which does not depend on DeepSeek's official service. When the official channel has problems, the backup is not affected by the same incident.
- Set the routing mode to failover, put the primary channel first, then use connectivity testing to confirm the backup channel is usable.
Failover is free and is not charged per token.
Frequently asked questions
DeepSeek won't open and reports that the server is busy. What can be done?
Check status.deepseek.com first to confirm whether this is an official outage. If it is, temporarily switching to another channel or another model is one option. For long-term use, backup channels are worth setting up in advance.
How is a current DeepSeek outage confirmed?
The official status page shows the chat service, the V4 Pro API, the V4.1 Flash API, the file upload service and the search service separately, and records the handling progress of each incident.
Does failover switch to a different model automatically?
No. Failover happens between channels that can serve the same model, and different models are never mixed. The model is specified by the caller in each request, and the gateway routes the request to a channel that can serve it.
Is it safe to put an API key in the gateway?
The real key is stored encrypted on the local machine and is never uploaded. Tools and projects use a virtual key, which can be revoked individually at any time.
Does failover cost anything?
No. The gateway endpoint, virtual keys, usage and cost statistics, automatic failover and local encryption are all part of the free features. Team features such as balanced routing and budget limits are part of VIP.
Where things stand now
The October 15 outage is resolved and compensation is being distributed. No one knows in advance when the next anomaly will occur. Connecting several channels for the same model to the gateway ahead of time keeps a project from coming to a halt when a failure lands.
AI Gateway is released with ServBay 1.33.0. Existing users can upgrade by clicking Check for Updates in Settings, and new users can download it at www.servbay.com/download.
Top comments (0)