A terminal agent that bets everything on DeepSeek
Whale gives DeepSeek a terminal-native home, skipping generic wrappers to squeeze value from 1M-token context and cheap caching.
What it does
Whale is a Go-based coding agent that lives in your terminal and talks exclusively to DeepSeek. It reads and edits files, runs shell commands, searches the web, and plugs into MCP servers. You can interact via a TUI for long sessions, fire one-shot questions from the CLI, or run headless for CI automation.
The interesting bit
Rather than chasing model-agnostic flexibility, Whale doubles down on DeepSeek-specific perks like 1M-token context and prompt caching. Its JavaScript workflow engine lets you script parallel agents and pipelines, and it claims to run Claude Code workflow scripts unchanged—a rare nod to portability in an otherwise opinionated tool.
Key highlights
- DeepSeek-native: optimized for 1M-token context, tool calling, and cost efficiency instead of supporting every LLM.
- Persistent sessions that survive for days; search history, branch, and resume where you left off.
- Three interfaces: interactive TUI, one-shot CLI, and headless mode for automation.
- Extensible via MCP servers, community skills, plugins, and lifecycle hooks.
- JavaScript workflow scripting for multi-agent fan-out, review pipelines, and adversarial validation.
Caveats
- Still in active development; the README recommends it mainly for personal or experimental repositories where changes can be reviewed and rolled back.
- Desktop and additional environments are listed as “on the way,” not yet available.
- Not affiliated with DeepSeek Inc.
Verdict
Terminal-dwelling developers who want a DeepSeek-optimized assistant with long memory and CI-friendly automation should try Whale. If you need a model-agnostic IDE replacement, look elsewhere—this is explicitly neither.