Evot
The lightest harness for agentic work.
An open-source terminal coding agent. A short prompt, four core tools, and the model does the thinking.
- 2026-11-03
/compute— add your ssh machines; the agent runs commands and long jobs on them for you. - 2026-10-18
Tab— context-aware suggestions, e.g. “Run the tests.”Tabcycles;→accepts without sending. - 2026-10-14
/loop— babysit a PR, poll a deploy, check back on a build; the agent picks when to wake and when to stop. - 2026-09-16
/task— scheduled runs with push delivery. - 2026-09-13
/share— read-only session links on evot.ai. - 2026-09-02
ctrl+b— background a running command and keep talking.
curl -fsSL https://evot.ai/install | shThe installer configures your PATH. Open a new terminal, or run the command it prints to use evot in the current terminal. Then log in:
evot loginLogin opens the TUI. Use hosted models or bring your own API keys.
Build from source
git clone https://github.com/evotai/evot.git
cd evot
make setup && make installBring your own models
Configure providers in ~/.evotai/evot.env. Examples:
# Anthropic
EVOT_LLM_ANTHROPIC_API_KEY=sk-ant-...
EVOT_LLM_ANTHROPIC_BASE_URL=your-anthropic-base-url
EVOT_LLM_ANTHROPIC_MODEL=claude-opus-4.8
# OpenAI Responses API
# EVOT_LLM_OPENAI_API_KEY=sk-...
# EVOT_LLM_OPENAI_MODEL=gpt-5.6-sol
# EVOT_LLM_OPENAI_PROTOCOL=openai_responses
# DeepSeek (Anthropic-compatible)
# EVOT_LLM_DEEPSEEK_API_KEY=sk-...
# EVOT_LLM_DEEPSEEK_BASE_URL=https://api.deepseek.com/anthropic
# EVOT_LLM_DEEPSEEK_PROTOCOL=anthropic
# EVOT_LLM_DEEPSEEK_MODEL=deepseek-v4-pro
# Kimi Coding (Anthropic-compatible)
# EVOT_LLM_KIMI_API_KEY=sk-...
# EVOT_LLM_KIMI_BASE_URL=https://api.kimi.com/coding
# EVOT_LLM_KIMI_PROTOCOL=anthropic
# EVOT_LLM_KIMI_MODEL=kimi-for-coding
# OpenRouter (Anthropic-compatible)
# EVOT_LLM_OPENROUTER_API_KEY=sk-or-...
# EVOT_LLM_OPENROUTER_BASE_URL=https://openrouter.ai/api/
# EVOT_LLM_OPENROUTER_PROTOCOL=anthropic
# EVOT_LLM_OPENROUTER_MODEL=stealth/ox-alphaUse comma-separated model names to configure multiple models. For OpenAI-compatible Chat Completions, set EVOT_LLM_OPENAI_PROTOCOL=openai and EVOT_LLM_OPENAI_BASE_URL to your endpoint. The official Responses API supports server-side context compaction, with local fallback.
Use evot from editors and other tools (ACP)
evot acp serves the Agent Client Protocol v1 on stdin/stdout, so any ACP client can drive evot:
evot acp # full coding tools
evot acp --tools readonly # read and search only
evot acp --tools none --ephemeral # answer-only, sessions kept in memorySupported methods: initialize, session/new, session/prompt, session/cancel, session/list, session/resume, session/close, and session/set_config_option for the per-session model selector. Tools run inside evot; it does not send session/request_permission. --tools none offers the model no tools, skills, or project instructions. --ephemeral writes no sessions to disk.
- Small by design. ~1k prompt tokens and four core tools:
read,bash,edit,write. - Less waiting. Long-running commands continue in the background while the agent works on independent tasks.
- Lean context. TypeSafe Jev prunes obsolete tool calls and trims stale results without summarising. Compaction is a fallback.
- Your choice of model. Free and low-cost hosted models, or your own provider keys.
Live performance & comparisons →
Apache-2.0