DeepAA CLI
Local gateway for coding agents: one setup for model providers, call tracing and per-request cost.
- 109 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The service stores provider keys and passes all agent traffic through itself.
- The install script is downloaded from the author's site and run in the shell.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/deepaaDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The service stores provider keys and passes all agent traffic through itself.
- The install script is downloaded from the author's site and run in the shell.
This is third-party code. Review the repository files before installing.
What it does
DeepAA runs on your machine between a coding agent and model providers. In one console you set up providers, models, keys and agents, then see what happened in a task: calls by task, subthread and step, prompts, answers, tool calls, tokens and time. It shows which tools, MCP servers, skills and rules entered each request's context. For every request cost is split into input, cache read and write and output, with prices, multipliers and discounts. For models you configure a chain of backups: on failure and before a response starts, the service switches to the next one. Five agents are integrated: Codex, Claude Code, OpenCode, DeepSeek Harness and ZCode. Accounting data stays local.
Who it is for. People using several coding agents and model providers who want to see spending and the causes of failures.
Good fit when
- You need to see why one Claude Code task burned so many tokens.
- Subscriptions and APIs from several providers must be combined into one ledger with balances and prices.
- A model needs a fallback chain in case a channel fails.
- You need to see which MCP servers and rules inflate request context.
Not a fit when
- You need a cloud observability service; DeepAA runs only locally.
- You work in an agent outside the list of five supported ones.
- Keys cannot be stored in an application on a work machine.
Example request
Show how many tokens and how much money my last Claude Code task used, and by which steps.Limitations
The interface and documentation are mostly Chinese, and an English README exists. The service stores provider keys and sees all agent traffic, so install it only on a personal machine. Five agents are supported, and available combinations depend on the protocol, model capabilities and agent version. Subscription cost is estimated by credit conversion formulas or proportionally to window usage when none exist, so treat it as an estimate. One-command install downloads a script from the author's site and installs Node.js if needed. From source you need Node.js 22.13 and pnpm. Fallback switching works only before the response starts arriving.
How to disable. Stop the service with `deepaa stop`, remove the service autostart and uninstall the package with `npm uninstall -g deepaa`. Delete keys in the OS credential store separately.
FAQ
How do I install DeepAA?
Run `curl -fsSL https://deepaa.dev/install.sh | bash` on macOS and Linux or `irm https://deepaa.dev/install.ps1 | iex` on Windows. Alternative: `npm install -g deepaa`.
How do I open the console after install?
`deepaa` starts the service and opens the console at `http://127.0.0.1:3210`. `deepaa status` and `deepaa stop` show state and stop the service.
Which agents are supported?
Codex, Claude Code, OpenCode, DeepSeek Harness and ZCode. You connect them in the providers section of the console after adding a model and key.
Does my data go to the cloud?
The author states that accounting and analysis data stay on your machine and are not sent to a DeepAA cloud. Requests go directly to the providers you configured.
How do I run DeepAA at login?
Run `deepaa service install`: the service is registered to start at boot and recover automatically after a crash.
Discussion
No questions or reviews yet.
Related
CLI toolsA self-hosted control center for coding agents: Claude Code, Codex, Copilot, OpenCode and Pi from your phone, desktop and terminal
ANOLISA is a system stack for AI agents: output compression, memory, workspace snapshots, command checks and token accounting
Shows what Claude Code and Codex actually send to the API: it captures local traffic and splits tokens and cost by source
A self-hosted gateway between AI apps and models, MCP tools and A2A agents: policy, budgets, routing and an audit trail
Official Dynatrace skills for agents: DQL, traces, logs, Kubernetes, cloud and security work inside Claude Code and Cursor.
A Docker sandbox for coding agents: a Debian image with Copilot CLI, Codex and Pi, and SSH, RDP and Docker enabled by flags.
README in short
DeepAA gathers provider setup, observation of agent calls and cost breakdown in one local workspace.