Evot
The lightest harness for agentic work.
An open-source coding agent for your terminal. A short prompt, four core tools, and the model does the thinking.
📢 News
- 2026-09-16
/taskputs the agent on a schedule — cron-timed runs with push delivery. - 2026-09-13
/shareturns your session into a read-only link on evot.ai. - 2026-09-11
GPT-5.6 Lunais free through Sep 18 🎉 — justevot login. - 2026-09-02
ctrl+bbackgrounds a long-running command so you can keep talking. - 2026-08-24 Free model of the week:
stealth/ox-alpha— free on OpenRouter for a week.
Less harness. More model.
- Lightweight. ~1k tokens, four core tools: read, bash, edit, write. You set the goal and constraints; the model chooses the path.
- Coordinated execution. Shells run in the background while the model advances independent work, waits for required results, and picks up completion notifications. Less idle time between thinking and doing.
- Affordable. Free and low-cost hosted models, or bring your own keys.
Jev prune
Instead of summarising old context, evot deletes what no longer matters. After each run a small judge model (TypeSafe Jev) works out which of your requests are still in play, then asks of every old tool call: does the task still depend on it, and will its result be read again? Calls that fail both go; stale results are truncated. Nothing is summarised, and compaction only runs if pruning alone is not enough.
Performance
Same task and environment, three agents × three models. Cost and tool calls—not wall-clock speed.
Task: fix a real bug in serde_json (issue #979) end to end.
All nine runs pass. In this eval, evot costs 72–78% less than Claude Code, with fewer tool calls on every model.
Latest models on the same task — full eval list on trace.evot.ai:
| Model | Eval |
|---|---|
| DeepSeek V4.1 Flash | run-210 |
| GPT-6 Astra | run-205 |
| Claude Fable 5.1 | run-202 |
| GLM-5.3 Flash | run-199 |
| stealth/ox-alpha | run-194 |
| DeepSeek V4 Pro | run-192 |
| GLM-5.3 | run-187 |
| Grok 4.6 | run-183 |
Installation
curl -fsSL https://evot.ai/install | shBuild from source
git clone https://github.com/evotai/evot.git cd evot make setup && make install
Login
evot login # follow the prompts; you land straight in the TUI after loginCustom configuration (bring your own models via ~/.evotai/evot.env)
# Anthropic (default) EVOT_LLM_ANTHROPIC_API_KEY=sk-ant-... EVOT_LLM_ANTHROPIC_BASE_URL=your-anthropic-base-url EVOT_LLM_ANTHROPIC_MODEL=claude-opus-4.8 # Multiple models: EVOT_LLM_ANTHROPIC_MODEL=claude-sonnet-5.0,claude-opus-4.8,claude-fable-5 # Or OpenAI Chat Completions # EVOT_LLM_OPENAI_API_KEY=sk-... # EVOT_LLM_OPENAI_BASE_URL=your-openai-compatible-base-url # EVOT_LLM_OPENAI_MODEL=gpt-5.6-sol # EVOT_LLM_OPENAI_PROTOCOL=openai # Or OpenAI Responses API (official OpenAI GPT/Codex models) # Using the official endpoint enables provider-native "remote compaction": # context is compacted server-side with far higher recall, taking priority # over the local algorithmic path (auto — falls back to local on any failure). # EVOT_LLM_OPENAI_API_KEY=sk-... # EVOT_LLM_OPENAI_MODEL=gpt-5.6-sol # EVOT_LLM_OPENAI_PROTOCOL=openai_responses # Or 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 # Or 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 # Or 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-alpha
License
Apache-2.0