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 /task puts the agent on a schedule — cron-timed runs with push delivery.
• 2026-09-13 /share turns your session into a read-only link on evot.ai.
• 2026-09-11 GPT-5.6 Luna is free through Sep 18 🎉 — just evot login.
• 2026-09-02 ctrl+b backgrounds 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:

Installation

curl -fsSL https://evot.ai/install | sh

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 login

# 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