Install.
Use npm, Homebrew, curl, Windows package managers, or source checkout.
picasso / docsindexed
Docs
Docs
Everything you need to run the agent well.
InstallUse npm, Homebrew, curl, Windows package managers, or source checkout.
AuthenticatePick managed mode, subscription passthrough, API keys, or local models.
RunStart the TUI, ask for a plan, approve the work, and inspect the result.
ModeWho pays inferenceBest forManaged.Picasso managed pool.Starting without a provider account.Subscription.Your Claude or ChatGPT plan.Users already paying providers.API key.Your provider account.Teams with token budgets.
First session.
Install the CLI. Connect one provider. Open a repository. Draft a plan. Run a verified edit.
Authenticate.
Pick managed mode, subscription passthrough, API keys, or local models.
Run.
Start the TUI, ask for a plan, approve the work, and inspect the result.
Reference.
CLI.
Every command family, common flags, output modes, and failure behavior.
Slash commands.
The interactive command language for plan, goal, memory, MCP, replay, and auth.
Hooks.
Eight extension points for shell or JavaScript handlers around tools, plans, sessions, and goals.
MCP.
Install, configure, sandbox, and serve external capability servers.
Generated reference.
The docs surface also publishes generated pages from repo docs and API exports for admin core, providers, auth, cache harness, control plane, goals, MCP, memory, sandbox, secrets, sessions, telemetry, tools, and TUI packages.
Source backed.
Generated docs are tied to checked-in package references instead of separate launch copy.
API exports.
Package APIs can be reviewed beside the product guides.
Trust material.
Security, telemetry, cache, and secret-handling references stay close to the code that implements them.
Guides.
First goal.
Write a completion contract, set budget limits, and watch the loop re-plan.
First memory profile.
Inspect, pin, edit, export, and delete memories after a real session.
Bring your subscription.
Use Claude or ChatGPT through Picasso without handing provider credentials to the cloud.
Customize agents.
Create named subagents with scoped tools and step budgets.
Choose your path.
| Mode | Who pays inference | Best for | Sponsor surfaces |
|---|---|---|---|
| Managed. | Picasso managed pool. | Starting without a provider account. | Dashboard and labeled terminal surfaces. |
| Subscription. | Your Claude or ChatGPT plan. | Users already paying providers. | Dashboard only. |
| API key. | Your provider account. | Teams with token budgets. | Dashboard only. |
| Local. | Your hardware. | Private Ollama or vLLM runs. | None required. |
Next.
Security review.
Read sandboxing, network policy, audit log, and sponsor boundaries.
Compare tools.
Pick between Claude Code, Codex CLI, Cursor, Devin, Copilot CLI, Aider, and Picasso.
Trust and legal.
Read data handling, privacy, terms, and sponsor boundaries.