Lispa Get started

A team of agents. A language of their own.

Your AI budget should
buy more progress.

Install the CLI →

Lispa’s agents develop their own language around the work—a compact way to use what they know, coordinate tasks, and get things done with fewer tokens.

Spend less on repeated explanations, oversized prompts, and routine steps that code or smaller models can handle. Save your best models for the problems that deserve them.

Your repo and toolsYour choice of modelsLocal or your cloud

Get started

Try it on your repo.

Create an account, then choose Lispa for a browser workspace or install the terminal agent below. Bring your model subscription or API keys.

Create an account

Already have one? Sign in.

  1. Install

    curl -LsSf https://lispa.ai/install.sh | sh -s -- --no-start

  2. Connect your account

    lispa setup

    Approve the terminal code in your browser.

  3. Open your repo

    cd ~/your-repo && lispa

    Use /login for a model subscription, or sync API keys during setup.

Existing agents, cloud hosts, and updates

Claude Code / Cursor: run lispa runtime install to connect the runtime and shared knowledge. Main-loop model routing stays with your existing agent.

Your cloud host: install on Linux, configure dependencies and credentials, and keep the process running in a persistent session. Hosting is separate; machines are not provisioned automatically.

Update: lispa update. Configure models with /tiers and inspect usage with /route. Provider usage is billed to your accounts.

Less to repeat

Keep what the work teaches.

Agents develop their own language of reusable operations. Findings stay available across sessions; large results stay out of the prompt until needed.

Less to spend

Match the model to the step.

Route routine work to cheaper models. Escalate when needed. Inspect the total cost, including reviews and retries.

Less to wait on

Share context. Work in parallel.

Delegated agents get the findings already collected. Run Lispa locally or on your own Linux cloud hosts, connected to the same workspace.

Behind the number

Same tasks.
Different model spend.

One development run: lower cost, longer runtime. Savings vary by task, model, and cache.

Evaluation details

Measured change: Lispa model routing enabled versus disabled. Routing v5 / executor v2, Opus session + GLM balanced tier, one repeat. Both runs scored 0.98 on deterministic checks. This measures model cost, not token reduction. Earlier runs saved little after review costs.