We use cookies to improve your experience, analyze site traffic and remember your preferences. By continuing to browse, you agree to our use of cookies.
Official terminal agent · Deeply bound to AIGS billing
AIGS Agent is a terminal AI coding agent with a fully transparent per-request token ledger. Strong models for the main loop, sensible tiers for subtasks, and cache reuse that keeps long sessions efficient.
$ aigs
main · ledger on · prompt cache on
> add email format validation to the login page
⚙ edit src/auth/login.tsx (+14 −2)
✓ lint · typecheck · test 12 passed
turn 3 · in 12,480 · cached 9,216 · 4.2s
Illustration: the main tier edits and verifies automatically, every turn lands in the ledger, context stays light
Capabilities
Core differentiators: cost transparency + cost awareness — usage, prices and progress, all in plain sight
Every request records usage, cache hits and cost — aggregated per turn, per session and per model, live in the status bar. Export a per-request CSV with /export cost and reconcile it line by line against the console usage page.
Strong models for the main loop to protect quality; compaction, summarization and other subtasks automatically pick more economical tiers, with the ledger recording where every token went.
Explicit prompt-cache management at the protocol layer plus an append-only context design keep hit rates high — repeated content is reused instead of recomputed.
Tool-result aging plus summary compaction. /compact with preview and undo keeps your context light without losing the plot.
After each change the agent runs lint / typecheck / test automatically and feeds failures back by file:line for up to three auto-fix rounds.
Readonly pass-through, bash allowlist with per-command approval, unconditional blocking of dangerous commands. Corporate proxies and self-signed certificates work out of the box.
Quick Start
Already have an AIGS account? One minute to wire up your terminal
Sign in to the console, create and copy your key (starts with sk-) on the API Keys page.
One command to install globally (requires Node.js 18+ or Bun).
npm i -g aigs-agent
Set the environment variable, then run aigs inside your project directory.
export AIGS_API_KEY=sk-your-key aigs
On Windows you can double-click start.cmd for a guided start; running from source also works (bun install && bun run dev).
Slash Commands
Type / in a session to open the command palette — these are the high-frequency ones
/costPer-request ledger and summary of this session
/export costExport a per-request CSV to reconcile with the console
/modelBrowse the live list and switch the main model
/model syncForce a platform sync (listing / delisting / prices)
/compactCompact context with preview and undo
/thinkingTune thinking effort to balance speed and usage
/permissionApproval mode with a per-command bash allowlist
/doctorSelf-check network and config — first stop for debugging
Visibility & Control
What you can see, you can control
Model, token usage, latency and cost of every request are on the record; export a full CSV with /export cost.
Per-task turn and spend caps are configurable — the agent pauses at the limit with progress preserved, never out of control.
The ledger reconciles line by line with the console usage page; prices sync with the platform automatically.
* Cache hits are billed at the platform cache price; rates vary by model — see the console for live pricing. Read the prompt caching billing guide
Cost-Aware
Built-in three-tier routing; the model list syncs with the platform automatically — listing, delisting and price changes apply in near real time
Complex tasks and code generation — quality first
Routine assistance — balanced
Compaction, summaries, titles — cost first
Browse available models and live prices →
Also using opencode, Claude Code, Codex or other third-party clients?
See the Agent integration guides →FAQ
AIGS Agent is built by AIGS and deeply bound to the platform's billing system: model lists and prices sync automatically, the per-request ledger reconciles line by line with the console usage page, and it is purpose-built for efficiency with cost-aware routing and cache optimization. opencode and Claude Code are general third-party clients — AIGS provides integration guides for them too.
Your API key stays on your machine (~/.aigs/config.json) and is never sent to any third party. Tool calls go through permission guards and approvals; sensitive files are write-protected by default and dangerous commands are always blocked.
The agent syncs the platform's live model list every 6 hours (listing, delisting and price changes apply automatically), and you can force a refresh anytime with /model sync. Delisted models are clearly flagged with guidance to switch.
Built-in dual adapters for OpenAI and Anthropic: GLM, DeepSeek, Qwen and more via the OpenAI-compatible endpoint; the Claude family via the native /v1/messages entry with discounted cache billing — no third-party LLM SDK involved.
It is a cross-platform CLI: macOS, Linux and Windows all work, requiring Node.js 18+ (recommended) or Bun. On Windows you can double-click start.cmd for a guided start.
Tool results age out automatically, and /compact does summary compaction — with preview to inspect the result, full for a deeper pass, and undo to roll it back in one step. Long sessions stay light.
Free credits on sign-up, one npm command, one minute to start
npm i -g aigs-agent
macOS / Linux / Windows · requires Node.js 18+ or Bun