Auto-Hydration
Pre-Hydrated Workspace & Git Context
Automatically hydrates `@file.ts#10-40` line ranges, prompt-referenced workspace files, and `<git_context>` before Step 1 — resolving repository Q&A in a single turn.
◆ AivoCode (`aivo`) v1.0.1·CLI Coding Agent & Harness
The Native CLI Coding Agent Built for AivoKey.
Purpose-built for AivoKey: Auto-Hydrated Context, 7 Specialist Agent Teams, 3-Wire Sticky Prompt Caching, and an Anti-AI-Slop Review Loop. Measured benchmark results are below.
9 downloads · 9 in the last 30 days
irm https://aivokey.dev/cli/download/install.ps1 | iexOne-line installer for Windows PowerShell. Requires Node.js >= 20.
artificial analysis coding agent index v2.0
AivoCode (`aivo`) Benchmark Results
Evaluated using the 3-pillar Artificial Analysis Coding Agent Index v2.0 methodology (FrontierCode v1.1, Terminal-Bench 4.0, and SWE-Atlas-QnA-AA) across 15 end-to-end software engineering tasks. The numbers below are AivoCode's own measured results.
Coding Agent Index v2.0
100.0%
15/15 tasks resolved · 0 reward hacks
API Cost / Task
$0.0063
Average across 15 tasks
Tokens / Task
28.7K
Average across 15 tasks
Median Wall Time
21.4s
3.5 steps on average
| Benchmark Pillar | Score | Wall Time | API Cost |
|---|---|---|---|
| FrontierCode v1.1 (Repo Patching · 5 Tasks) | 100.0% | 33.4s (3.8 steps) | $0.0074 (30.4K tok) |
| Terminal-Bench 4.0 (Terminal Workflow · 5 Tasks) | 100.0% | 22.6s (5.6 steps) | $0.0050 (37.1K tok) |
| SWE-Atlas-QnA-AA (Repository Q&A · 5 Tasks) | 100.0% | 8.2s (1.0 step) | $0.0066 (18.7K tok) |
| Composite Index (Coding Agent Index v2.0 · 15 Tasks) | 100.0% | 21.4s (3.5 steps) | $0.0063 (28.7K tok) |
changelog
AivoCode Release Notes
Newest first. Every stable release can be downloaded directly from this page.
- v1.0.1Oct 28, 2026
Client identity headers and a version command
- Requests from AivoCode carry a client identity header, so AivoKey reports AivoCode usage per model in the dashboard.
- -v, --version, and the new aivo version command print the installed version.
- The release ships as aivocode-1.0.1.tgz; the 1.0.0 tarball stays available.
- v1.0.0Oct 28, 2026
First stable release
- First stable release of the AivoCode CLI.
version 1.0 harness architecture
AivoCode v1.0 Coding Harness Capabilities
3-Wire Engine
Multi-Wire Protocol, Sticky Cache & Stall Recovery
Speaks `/v1/chat/completions`, `/v1/messages`, and `/v1/responses` natively with deterministic system prompts for 90% prompt cache discounts, adaptive step reasoning effort, and automatic stream-stall failover.
7 Agent Teams
7 Specialist Agent Teams & Multi-Stage Workflows
Built-in multi-role engineering pipelines (`[ARCH → BACK → FRONT → QA]`): `fullstack`, `backend`, `frontend`, `android`, `blender3d`, `gamedev`, and `autopilot`.
Anti-AI-Slop
Iterative Review → Fix Loop & Turn Checkpoints
Audits modified code against lazy `// TODO` stubs, truncated implementations, and blind `any` casts, triggers a dedicated Fixer agent pass when needed, and snapshots every turn for `/undo` and `/redo`.
Cross-Shell
Cross-Shell Normalization & Non-Interactive Git
Automatically normalizes `&&`, `||`, `2>/dev/null`, `ls -la`, `rm -rf`, and `del` on Windows PowerShell with non-interactive Git environment variables so agent shell commands succeed on the first try.
Subagent Routing
Plan-Aware Model Filtering & Fast Subagents
Cross-references `/v1/models` and `/v1/key` to prevent `403 model_not_in_plan` errors, and delegates read-only `explore` subagents to fast/free-token models while keeping complex edits on your primary model.
Zero-Key Search
Client-Side WebSearch, CodeSearch & WebFetch
Built-in `web_search`, `code_search`, and `web_fetch` powered by keyless Exa MCP & Parallel MCP (plus Tavily, Firecrawl, Brave, and DuckDuckGo failover) — zero server search surcharges.
Self-Evolving
Self-Learning Memory & Multi-Harness Skills
Recalls project architecture and bug lessons from `~/.aivokey/memory` without breaking prompt cache prefixes, loads skills from `.aivokey`, `.opencode`, and `.claude`, and creates reusable skills via `skill_create`.
quick documentation
Quick Documentation & CLI Reference
Core subcommands and flags for running AivoCode v1.0 in local repositories or headless CI/CD pipelines.
CLI SUBCOMMANDS
aivoStart an interactive coding agent session in the current workspace
aivo login sk-aivo_...Verify your AivoKey API key and configure default primary & fast subagent models
aivo run "<prompt>" [--json]Run a coding task non-interactively (alias: `aivo -p "<prompt>"`)
aivo workflow "<goal>" --team <id>Execute a multi-stage Specialist Agent Team workflow (`fullstack`, `backend`, `frontend`, `android`, `blender3d`, `gamedev`, `autopilot`)
aivo models [--tools|--free]List live AivoKey models, prices per 1M tokens, capabilities & account eligibility
aivo status / aivo doctorInspect live PAYG balance, Plan 5h/7d/month windows, free tokens & gateway connectivity
aivo sessionsList saved JSONL sessions with message counts and cumulative micro-USD cost (`-c` to resume)
HARNESS FLAGS & OPTIONS
-m, --model <id>Canonical model ID (e.g. `deepseek-v4.1-flash`, `claude-opus-5-5`, `gpt-6.1-sol`)
--fast-model <id>Fast model used for read-only `explore` subagents
--effort <level>Reasoning effort level (`off`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max`)
--wire <protocol>AivoKey wire protocol (`chat`, `messages`, `responses`)
--mode <mode>Execution permission mode: `auto` (build), `plan` (read-only), or `safe` (confirm)
-t, --team <id>Specialist agent team (`fullstack`, `backend`, `frontend`, `android`, `blender3d`, `gamedev`, `autopilot`)
-c / -s <id>Continue the most recent session (`-c`) or resume a specific session ID (`-s`)
Ready to build with AivoCode v1.0.1?
Create an API key in the console, claim free tokens on the faucet, and run `aivo login`.