- type
- Coding agent
- interface
- Terminal
- openSource
- Yes (MIT)
- lineage
- Replaces a sequential tool-call loop with executable DAG-based 'graph calls' mixing normal tool calls with calls to Jev, TypeSafe AI's separately-vendored System One model (a fast, non-generative, typed-decision model, not an LLM), reserving slow System Two LLM reasoning for the decisions that need it; its own README benchmark table reports, e.g., 128 tool calls / 16 LLM calls / 50 Jev calls in 3m26s on a conversation_eval task, versus 60+ LLM calls for comparison agents on the same task (vendor self-reported, not independently verified).
- note
- GitHub repo merijjeyn/jive created 2026-09-21 (confirmed via GitHub API: 2026-09-21T15:07:55Z), inside this lens's window (2026-09-19 to 2026-09-28); surfaced via GitHub repository search, not a Show HN post. MIT licensed (confirmed via GitHub API), 160 stars at time of check. The README describes Jive as a terminal coding agent that plans work as executable graphs and pairs tool calls with calls to Jev; it demonstrates task completion via linked benchmark videos, consistent with real code editing and command execution rather than planning chat only. Supported LLM providers are not stated in available sources beyond its use of Jev. The benchmark figures for this entry (128 tool calls, 16 LLM calls, 50 Jev calls in 3m26s on conversation_eval, versus 59 tool/60 LLM calls for Codex and 97 tool/98 LLM calls for Claude Code on the same task) come from the project's own README table and are vendor self-reported, not independently verified; the table itself carries no verification disclaimer. Distinct from the catalogue's existing dual-model precedents, Devin Desktop's Fusion (frontier lead model paired with a cheaper execution sidekick) and Reasonix's executor and planner split, both of which pair two LLMs: Jive instead pairs an LLM with a non-generative typed-decision model, TypeSafe AI's Jev, a System One typed-decision model, inside an executable graph, cutting LLM calls rather than swapping which LLM does the work.
