ais

Scriptable, multi-provider AI agent session manager.
Create, monitor, and orchestrate Claude Code and Kimi Code sessions in tmux.

Why ais?

Most multi-agent tools give you a dashboard. ais gives you a scriptable CLI that agents themselves can operate — enabling agent-orchestrates-agents workflows.

ais is built for power users who run fleets of AI agents, rotate across accounts to avoid rate limits, and want their orchestrator agent to manage the whole thing.

Quick Start

Install

# Option A: Clone and add to PATH
git clone https://github.com/gastown-publish/ais.git
cp ais/bin/ais ~/.local/bin/ais
chmod +x ~/.local/bin/ais

# Option B: Direct download
curl -sL https://raw.githubusercontent.com/gastown-publish/ais/main/bin/ais \
  -o ~/.local/bin/ais && chmod +x ~/.local/bin/ais

Make sure ~/.local/bin is in your PATH.

Set up an account

# Claude Code
mkdir -p ~/.claude-accounts/cc1
CLAUDE_CONFIG_DIR=~/.claude-accounts/cc1 claude auth login

# Kimi Code (free)
mkdir -p ~/.kimi-accounts/kimi1
KIMI_SHARE_DIR=~/.kimi-accounts/kimi1 kimi login

Run your first agent

# See your accounts
ais accounts

# Launch an agent
ais create worker1 -a claude -A cc1 --yolo -c "fix the auth bug"

# Check progress
ais inspect worker1 -n 20

# Watch it work
ais watch worker1

  NAME                 AGENT    ACCOUNT    DIR                            AGE
  ──────────────────── ──────── ────────── ────────────────────────────── ────────
  worker1              claude   cc1        ~/project                      12m
  worker2              claude   cc2        ~/project                      8m
  kimi-refactor        kimi     2          ~/other-project                3m

  Claude Code accounts (~/.claude-accounts)
  ─────────────────────────────────────
    cc1           (logged in)
    cc2           (logged in)
    cc3           (configured)

  Kimi Code accounts (~/.kimi-accounts)
  ─────────────────────────────────────
    1             (API key)
    2             (OAuth)

Real-World Workflows

Parallel bug hunt across 3 agents

# Spin up 3 agents on different accounts to avoid rate limits
ais create hunt1 -a claude -A cc1 -d ~/app -c "find the memory leak in src/server/"
ais create hunt2 -a claude -A cc2 -d ~/app -c "audit error handling in src/api/"
ais create hunt3 -a kimi   -A 1  -d ~/app -c "review database queries for N+1 issues"

# Monitor all of them
ais ls

# Check on progress
ais inspect hunt1 -n 100

# When done, clean up
ais kill --all --save

Credit rotation

# Check remaining credits across all accounts
./scripts/check-credit.sh      # Claude
./scripts/check-kimi-credit.sh  # Kimi

# Check specific accounts only
./scripts/check-credit.sh cc1 cc3

╔══════════════════════════════════════════════════════════════════════╗
║  Claude Code Credit Check (GMT+7)                                  ║
╚══════════════════════════════════════════════════════════════════════╝

┌──────────────┬──────────┬──────────┬───────────┬──────────────────────────┐
│ Account      │ Week All │  Sonnet  │ Remaining │ Resets (GMT+7)           │
├──────────────┼──────────┼──────────┼───────────┼──────────────────────────┤
│ cc1          │    6%    │    0%    │   94% OK  │ Mar 06, 10:00am (GMT+7)  │
│ cc2          │    1%    │    0%    │   99% OK  │ Mar 06, 12:00pm (GMT+7)  │
│ cc3          │    2%    │    0%    │   98% OK  │ Mar 06, 11:00am (GMT+7)  │
└──────────────┴──────────┴──────────┴───────────┴──────────────────────────┘

  Best account: cc2 (99% remaining)

# Switch to a fresh account when one hits limits
ais kill worker1
ais create worker1 -a claude -A cc2 -d ~/app -c "continue from where cc1 left off"

Agent-orchestrates-agents (via OpenClaw)

The skills/ais-orchestrator/SKILL.md enables an OpenClaw agent to:

1. Spawn worker agents with ais create
2. Monitor their output with ais inspect
3. Detect rate limits and rotate accounts
4. Collect results and coordinate across sessions

This is the "AI that runs AIs" pattern — your orchestrator agent manages a fleet of coding agents without human intervention.

Usage Reference

Create a session

ais create <name> [options]

Options:
  -a, --agent TYPE      Agent: claude, kimi (default: kimi)
  -A, --account ID      Account: cc1..nicxxx for claude, 1..N for kimi
  -c, --cmd TEXT        Command to inject after agent loads
  -d, --dir PATH        Working directory (default: current)
  --yolo                Auto-approve mode
  --attach              Attach to session after creation
  --size WxH            Terminal size (default: 160x50)
  --                    Pass remaining flags to agent CLI

Manage sessions

ais ls                          # List all managed sessions
ais inspect <name> -n 200       # Capture last 200 lines of output
ais inspect <name> --rate-limit # Check for rate limit errors
ais inject <name> "run tests"   # Send command to session
ais watch <name> -i 5           # Watch live, refresh every 5s
ais watch <name> --until "done" # Watch until pattern appears
ais logs <name>                 # Save full scrollback to file
ais status <name>               # Machine-readable status + log trail
ais kill <name>                 # Graceful shutdown
ais kill <name> --save          # Save logs, then kill
ais kill --all --force          # Force kill everything

Log trail & agent follow-up

Every session creates a log trail at ~/.ais/logs/<session>/:

~/.ais/logs/worker1/
├── events.log      # Timestamped event log
├── snapshots.log   # TUI captures at each analysis step
├── status.json     # Machine-readable status (for other agents)
└── prompt.txt      # Original task prompt

Other agents can pick up where a session left off:

# Read the original prompt
cat ~/.ais/logs/worker1/prompt.txt

# Check machine-readable status
ais status worker1
# {"session":"worker1","status":"running","detail":"command injected","updated":"..."}

# Resume or retry
ais kill worker1
ais create worker1 -a claude -A cc2 -d ~/app -c "$(cat ~/.ais/logs/worker1/prompt.txt)"

Tools Included

Account Setup

Each agent account gets its own isolated directory with separate credentials, so you can run multiple agents simultaneously without conflicts.

Claude Code (quick setup)

# 1. Create account directory
mkdir -p ~/.claude-accounts/cc1

# 2. Log in (opens browser, paste code back)
CLAUDE_CONFIG_DIR=~/.claude-accounts/cc1 claude auth login

# 3. Auto-approve permissions (ais does this automatically, but you can do it manually)
cat > ~/.claude-accounts/cc1/settings.json << 'EOF'
{
  "permissions": {
    "allow": ["Bash(*)", "Read(*)", "Write(*)", "Edit(*)", "Glob(*)", "Grep(*)", "WebFetch(*)", "WebSearch(*)"],
    "deny": []
  }
}
EOF

Running as root? Prefix with IS_SANDBOX=1. Inside another Claude session? Add CLAUDECODE=.

Repeat for each account (cc2, cc3, etc.) — each needs its own separate OAuth login.

Kimi Code (quick setup)

# Option A: OAuth (free)
mkdir -p ~/.kimi-accounts/kimi1
KIMI_SHARE_DIR=~/.kimi-accounts/kimi1 kimi login

# Option B: API key (from platform.kimi.com)
mkdir -p ~/.kimi-accounts/kimi1
cat > ~/.kimi-accounts/kimi1/config.toml << EOF
[providers."managed:kimi-code"]
type = "kimi"
base_url = "https://api.kimi.com/coding/v1"
api_key = "sk-YOUR-KEY"
EOF

Verify

ais accounts    # Should list all configured accounts

See docs/account-setup.md for detailed instructions, container setup, and troubleshooting.

Architecture

ais uses tmux as the execution layer. Each agent session runs in a named tmux session with metadata stored in environment variables:

The orchestrator skill (skills/ais-orchestrator/SKILL.md) provides a higher-level interface for OpenClaw agents to coordinate multiple sessions — handling task distribution, rate limit rotation, and result collection.

Prerequisites

• bash 4.0+
• tmux 3.0+
• Claude Code CLI (~/.local/bin/claude) and/or Kimi Code CLI (~/.local/bin/kimi)
• At least one configured account in ~/.claude-accounts/ or ~/.kimi-accounts/

FAQ

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

This project uses beads for issue tracking:

bd ready          # Find available work
bd show <id>      # View issue details
bd close <id>     # Complete work

See AGENTS.md for the agent collaboration workflow.

Prior Art

This repo consolidates tools originally shared as standalone gists:

• Claude credit checker + agent setup — check-credit.sh, ais, orchestrator skill, agent setup guide
• Kimi credit checker + multi-account — check-kimi-credit.sh, kimi-account, Kimi multi-account README

This repo is the canonical, maintained version. The gists remain for reference.

License

MIT — see LICENSE.