DeepSeek in OpenCode: Excession provider setup
Configure DeepSeek in OpenCode with an Excession custom provider, an environment-based API key and a tested Chat Completions request.
Research and product notes
Security research, API guides and updates from Excession Labs.
Subscribe via AtomDeepSeek in Claude Code: Excession gateway setup
Configure DeepSeek in Claude Code through the Excession Anthropic-compatible gateway, with a tested client configuration and its compatibility limits.
What are abliterated models?
What abliterated models are, how they differ from uncensored fine-tunes, what can change in quality, and when to use a hosted API or model download.
Excession API vs OpenRouter: integration and funding
A sourced comparison of SDK setup, separate model catalogs, routing, credit purchases and expiration.
API authorization testing: a reproducible local workflow
Run a localhost-only authorization fixture, reproduce cross-account access and verify a server-side fix with recorded curl evidence.
DeepSeek API latency: timings, errors and raw data
40 measured requests: first-token versus visible-answer latency, errors, raw JSON/CSV and a script to repeat the test.
DeepSeek API guide: pricing, Python, streaming and tool calls
Choose a model, create a key and send your first request. A practical integration guide with model-specific prices, streaming, complete tool-call handling and error recovery.
-
The alleged Anthropic Numbat leak is likely fake
The alleged Anthropic Numbat leak on Hugging Face is likely fake: parts of the model match randomly generated numbers exactly, so they were never trained. We tested it in 90 KB without running its code.
-
Introducing Incursion
Incursion by Excession Labs is a terminal agent for authorized offensive security work. A planner splits the job, workers run in parallel, and each finding comes back with its evidence. This post tells what it does and how to start.
-
Simple pricing: pay as you go, or a plan with up to 150% more
Add funds from $0.50 that never expire while your account is open, or pick a plan that gives 50% to 150% more plan credit than you pay. Both use the same per-token prices, and a plan lowers your effective price per token.