Devin Hits the Terminal, Multi-Model
Cognition just dropped Devin for Terminal. The pitch — a CLI agent that keeps working when you close your laptop. Start a session locally, hand it off to the cloud, walk away. Come back to a PR.
The model lineup is the interesting part. Opus 4.7. GPT-5.5. And Cognition's own SWE-1.6. Multi-model out of the box, not locked into any one lab's tokens. That's a direct shot at Claude Code (Anthropic-only) and Codex (OpenAI-only). Cursor Background and Devin's web dashboard already had this idea — Devin for Terminal collapses it to one CLI command.
The setup matters because Cognition has been operator-mode-pilled since Devin 1.0 in 2024. They didn't pivot — everyone else slowly caught up. The current coding-agent stack (Claude Code, Codex, Windsurf, Cursor) is still mostly assistant-mode: you sit and wait while it works. Cognition's whole thesis is the laptop close. Run the agent in a sandbox far away from your machine, sleep, come back to a PR. They were early. Now the rest of the stack is catching up to them and Cognition is using a CLI to make sure it stays in front.
What's missing from the launch — concrete pricing tiers (a free option exists but enterprise pricing isn't disclosed), how the cloud handoff handles model switching mid-task, and whether Anthropic and OpenAI are happy about a third-party CLI multi-tenanting their flagship models. The bigger question — when every shop's agent CLI runs on the same three frontier models, the moat is the harness. Cognition has been building harness longer than anyone.
Devin for Terminal is on Product Hunt today: https://www.producthunt.com/products/devin-1-2-by-congition
← Back to all articles
The model lineup is the interesting part. Opus 4.7. GPT-5.5. And Cognition's own SWE-1.6. Multi-model out of the box, not locked into any one lab's tokens. That's a direct shot at Claude Code (Anthropic-only) and Codex (OpenAI-only). Cursor Background and Devin's web dashboard already had this idea — Devin for Terminal collapses it to one CLI command.
The setup matters because Cognition has been operator-mode-pilled since Devin 1.0 in 2024. They didn't pivot — everyone else slowly caught up. The current coding-agent stack (Claude Code, Codex, Windsurf, Cursor) is still mostly assistant-mode: you sit and wait while it works. Cognition's whole thesis is the laptop close. Run the agent in a sandbox far away from your machine, sleep, come back to a PR. They were early. Now the rest of the stack is catching up to them and Cognition is using a CLI to make sure it stays in front.
What's missing from the launch — concrete pricing tiers (a free option exists but enterprise pricing isn't disclosed), how the cloud handoff handles model switching mid-task, and whether Anthropic and OpenAI are happy about a third-party CLI multi-tenanting their flagship models. The bigger question — when every shop's agent CLI runs on the same three frontier models, the moat is the harness. Cognition has been building harness longer than anyone.
Devin for Terminal is on Product Hunt today: https://www.producthunt.com/products/devin-1-2-by-congition
Comments