Questions worth answering plainly.
Picasso is a coding-agent CLI for real software work. It is free through the managed path, sponsor-supported by design, and built with visible boundaries.
Picasso is a coding-agent CLI for real software work. It is free through the managed path, sponsor-supported by design, and built with visible boundaries.
Product.
What is Picasso.
Picasso is a coding-agent CLI. It runs in the terminal, reads a repo, understands context, plans changes, edits files, runs commands, writes tests, debugs failures, explains its work, and keeps state across sessions.
Is Picasso a chat box for code.
No. Picasso is a full agent harness with a plan, act, observe, reflect loop, Plan mode, persistent memory, `/goal`, subagents, MCP support, provider routing, permission controls, and the Studio.
What is the Studio.
The Studio is Picasso's companion dashboard for memory, plans, goals, sessions, replay, account state, credential modes, and managed usage visibility.
What is a canvas.
A canvas is a Picasso session. It can persist, resume, fork, replay, and sync to the Studio.
Pricing and sponsors.
Is Picasso really free.
Yes, for individual developers through the managed path. Picasso pays for inference in that mode through sponsor support.
How is Picasso paid for.
Companies sponsor access for builders. Sponsors appear in labeled surfaces, mainly in the Studio, and in a one-line terminal disclosure marked `◇ Sponsored`.
Where do sponsor messages appear.
Primarily in the Studio, around account state, memory, sessions, plans, goals, and replay. In the terminal, the sponsor surface is one line, dismissible, and clearly marked.
Do sponsors influence code or model output.
No. Sponsors do not affect generated code, model output, model routing, tool dispatch, memory retrieval, or subagent selection.
Can I disable terminal sponsor surfaces.
The terminal sponsor disclosure is designed to be dismissible and clearly marked. Paranoid mode can hide terminal sponsor surfaces.
Models and credentials.
Which models can I use.
Picasso supports routing across Anthropic, OpenAI, Microsoft Foundry, Google, xAI, Qwen, Xiaomi MiMo, Kimi, DeepSeek, Zhipu GLM, MiniMax, OpenRouter, Ollama, and vLLM.
Can I bring my own API key.
Yes. BYO API key mode lets you use your own provider credentials and keep billing with that provider.
Can I use my Claude or ChatGPT subscription.
Yes, where provider terms and account configuration allow subscription passthrough.
Can I run local models.
Yes. Picasso supports local model paths through Ollama and vLLM.
Does Picasso work offline.
Picasso can work with local models and local repo context when configured for local use. Features that depend on hosted providers, the Studio, sync, or remote MCP servers require network access.
What happens when a budget is hit.
Picasso stops or asks for direction at the boundary. In `/goal`, it continues until completion criteria pass or a budget, approval, or permission boundary is reached.
Memory and data.
Does Picasso train on my code.
No. Picasso memory is retrieval only. Your code is not used for training by Picasso memory.
What does memory store.
Memory can store codebase facts, style preferences, decision history, commands, constraints, and other context that helps the agent work better in future sessions.
Can I see memory.
Yes. Memory is user-visible in the Studio.
Can I edit, export, or delete memory.
Yes. Memory is editable, exportable, and deletable.
Safety and workflows.
Is this safe for production repos.
Picasso is built for production work, but it still works under your rules. It supports workspace containment, approval gates, shell allowlists and denylists, network controls, secret detection, audit logs, and permission modes.
What is `/goal`.
`/goal` is Picasso's mode for long-running autonomous work. You give it a target, completion criteria, and a budget. Picasso keeps working until the criteria pass or a boundary is reached.
What are subagents.
Subagents are parallel brushes with named roles and isolated context. They can inspect, implement, verify, or research separate parts of a task while the main session stays in control.
Can teams use it.
Yes. Picasso works for individuals and teams, especially where shared plans, replay, memory, and visible sessions matter.