What happened
Help Net Security’s November 2, 2026 review describes Byoky as an open-source AI API-key wallet distributed as Chrome and Firefox extensions and as iOS and Android apps. Users add keys for OpenAI, Anthropic, Google Gemini and other providers. The keys are encrypted on the user’s device; an app requesting access receives a session token, while Byoky sends the request to the provider. The stated goal is to avoid pasting a provider key into each app’s settings and losing control over where it is stored.
The review says users set a master password of at least 12 characters. Byoky encrypts the vault with AES-256-GCM and derives the encryption key using 600,000 rounds of PBKDF2. Developers can continue using official Anthropic or OpenAI libraries, passing them a session key and a fetch function from Byoky’s SDK. The documentation says streaming, tool calls and image inputs are supported. Users can set per-app spending caps, and a local log records each request’s app origin, provider and status.
The wallet supports 13 cloud providers as well as local Ollama and LM Studio servers. It can translate request formats among Anthropic, OpenAI, Gemini and Cohere, which the review says lets a Claude-oriented app use a GPT key without code changes by its developer. Command-line agents including OpenClaw, Claude Code and Hermes connect through a local proxy called the bridge. The bridge requires a bridge token, intended to prevent other programs on the same machine from spending the keys.
The review highlights a different bridge behavior: when a third-party agent runs with a Claude Pro or Max subscription token, the bridge renames the agent’s tools and moves its system prompt so that Anthropic’s first-party detection accepts the request. The source characterizes this as a less-advertised behavior. It does not establish how Anthropic interprets this specific implementation or whether any individual use has been challenged. The review’s accompanying account says the latest GitHub release is v1.1.1, dated October 25, 2026, and that v1.1.0 removed cloud sync.
What the archive adds
An earlier Telegram post described a Claude Code proxy that redirected requests to alternative providers, and another described running GPT-5.6 Sol inside Claude Code through a local proxy. Those posts are related by the use of a proxy to change what sits behind a Claude Code workflow; this is a new product and a distinct reported behavior, not a restatement. The archive has no directly related Field Note.
Trend signal
Byoky’s key custody and per-app permissions fit the idea that agent-ready software exposes bounded, permissioned capabilities. The review describes apps asking a wallet for access rather than receiving a raw key, a concrete example of that product boundary: Agent-ready software exposes capabilities. The reported subscription-detection evasion complicates the picture: permissioning at the wallet layer does not itself establish permission under a provider’s subscription terms.
Why now
The review was published November 2, 2026, shortly after the reported v1.1.1 release on October 25. It also says v1.1.0 removed cloud sync, making on-device custody especially salient to the product’s current configuration. The source does not explain why the bridge behavior is newly being scrutinized.
Why it matters
For teams, the wallet addresses a real operational problem: API keys copied into many applications are difficult to inventory and revoke. A broker that issues session access, limits spending by app and keeps a local request log can make that access more legible and easier to constrain. Those controls are useful only if teams also understand the trust boundary: who can invoke the bridge, how the session is scoped, and what the local token protects against.
The reported Claude subscription behavior is a separate risk, not a feature to treat as ordinary key management. A bridge can technically make an agent’s requests appear acceptable to a detection system without establishing that the subscription permits the agent’s use. Teams should distinguish API credentials, which are intended for programmatic access under their own terms, from subscription tokens tied to a first-party client. Before adopting any proxy that modifies requests for detection purposes, confirm provider rules and avoid routing work through a path whose compliance status is unclear.
Limits and uncertainty
The source is a product review, not an independent security audit. The supplied material does not report penetration testing, a threat model, key recovery details, how the bridge token is protected, or whether spending caps can be bypassed. It also does not establish that the reported behavior has been confirmed by Anthropic, that Anthropic has taken action, or how subscription terms apply to each configuration. Encryption parameters alone cannot show that the full product is secure. The report’s claims should be treated as descriptions of the reviewed product, not as proof of effectiveness or policy approval.
Skill to build now
Inventory where your team stores model credentials this week. For each app or agent, record whether it uses an API key or a subscription token, which process can access it, what spending limit applies, and how access is revoked. Treat proxy-based access as a separate integration requiring an explicit check of provider terms; do not infer permission from the fact that a request succeeds.
Reader takeaway
Byoky presents a practical alternative to distributing raw AI API keys across apps, with on-device storage, session access and spending controls. But its reported Claude bridge behavior crosses from credential management into detection evasion. Evaluate the two questions separately: whether a wallet protects keys, and whether a particular subscription permits the agent workflow being routed through it.