Orcus Launches Web-Credential AI SDK

// 1h agoOPENSOURCE RELEASE

Orcus Launches Web-Credential AI SDK

Orcus provides TypeScript SDKs that turn supported AI provider websites into API-style interfaces using authenticated web credentials instead of LLM API keys. Its Gemini package supports streaming, multimodal input, tool calling, structured outputs, and batch processing.

// ANALYSIS

Orcus offers a compelling way to prototype AI integrations without separate API billing, but its web-session approach trades setup convenience for upstream fragility and credential risk.

–The SDK gives developers familiar primitives without requiring a proxy or hosted gateway.
–Gemini is available now, while DeepSeek and ChatGPT integrations remain planned.
–Web interfaces can change without notice, making compatibility and long-term reliability uncertain.
–Provider credentials must be protected like secrets, and usage may still be constrained by account limits or terms of service.
–The MIT-licensed package could become a useful bridge for experimentation, especially where official APIs are unavailable or expensive.

// TAGS

orcusllmsdkapistreamingtool-usestructured-outputopen-source

DISCOVERED

1h ago

2026-10-24

PUBLISHED

2h ago

2026-10-24

RELEVANCE

8/ 10

AUTHOR

i_sphinx_

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

INFRA21m ago

boat Cuts Whisper Costs More Than 10x

boat provides persistent, full Ubuntu cloud VMs that let developers run Whisper transcription workloads on their own infrastructure. A user reportedly reduced transcription costs by more than 10x by moving from managed APIs to boat’s usage-based virtual machines.

SECURITY38m ago

Daybreak aids OpenCode breach response

OpenAI Daybreak Blue gives approved defenders reduced-refusal access to supported frontier models for incident response, secure code review, malware analysis, and patch validation. OpenCode reportedly turned to the program after its breach investigation encountered refusal barriers, exposing the tension between cyber safeguards and usable defense tooling. [OpenAI](https://help.openai.com/en/articles/20001258-openai-daybreak-trusted-access-for-cyber) [Axios](https://www.axios.com/2026/10/21/when-ai-models-refuse-to-fight-hackers)

RESEARCH2h ago

Claude Proves Security Across 123 Libraries

Anthropic and Theorem used Claude to generate Rocq-checked, machine-verifiable no-crash proofs for 239 entry points across 123 OSS-Fuzz libraries, operating on compiled RISC-V binaries. The process also uncovered bugs that had survived trillions of fuzzing runs. [Anthropic research](https://www.anthropic.com/research/formal-verification)