PAPERCLIP ADAPTER
External adapter plugin that runs Mistral Vibe agents — over ACP (vibe-acp) or headless via the vibe CLI. Each agent gets its own Vibe state and its own Mistral API key, so agents run side by side without colliding.
Made with Le Chonk · Mistral Large 4 (preview)
WHAT YOU GET
Every agent gets its own VIBE_HOME and its own Mistral API key. Config, sessions, caches and history never collide — concurrent agents can't corrupt each other.
Pick per agent: interactive ACP (vibe-acp) with approvals in the Paperclip UI, or headless vibe -p runs with policy-based approvals.
listModels / refreshModels query the Mistral API and merge with a curated static list, so the model picker always works — even offline.
Paperclip's skill library symlinks into the agent's own $VIBE_HOME/skills/. Self-healing on every run; user-installed skills stay untouched.
Optional "Vibe monthly spend" window for Paperclip's quota dashboard via the Mistral Admin API (Enterprise-only). Absent when not configured.
The ACP lane delegates to Paperclip's shared ACP engine; the CLI lane maps the vibe NDJSON stream to the run transcript. MIT licensed.
EXECUTION ENGINES
vibe-acp binary via Paperclip's shared ACP engine.vibe -p process per run.cliAgent profile (default auto-approve).--resume, stored per agent in VIBE_HOME; failed resume falls back to a fresh session.export.json.GET STARTED
In Paperclip: Settings → Adapters → Install from npm, then enter the package name.
paperclip-adapter-mistral-vibe
Install the vibe CLI (registers vibe-acp side by side) or download the standalone vibe-acp binary from the mistral-vibe releases. Create a Mistral API key under Code › Vibe CLI in the Mistral console.
uv tool install mistral-vibe
Add agent → Mistral Vibe: pick the execution engine and model, put your key in the agent env, then run Test environment.
MISTRAL_API_KEY=<your key>