Not live yet. The most beautiful coding agent ever made is almost here — and it's free.Join the waitlist →
Blog

Blog.

A few notes from the people building Picasso. What we believe. What we shipped. What we refuse to hide.

picasso / blogindexed
Blog
Blog

A few notes from the people building Picasso. What we believe. What we shipped. What we refuse to hide.

Code FreelyThe first choice in a developer tool is not the model. It is the door. Picasso starts at zero because the best way to understand an agent is to let it touch a real repository. Managed inference is paid for by labeled sponsor surfaces. Never by selling code. Never by sending prompts to ad systems. The exchange is visible because trust starts where the bill would normally appear.
Memory should have a windowMost agents forget. Some remember invisibly. Picasso takes the harder path: memory that can be inspected, edited, exported, and cleared. The point is not to make the agent feel human. The point is to stop repeating yourself while keeping control of what comes back.
The loop is the interfacePlan, act, observe, reflect. That is not internal plumbing. It is the product. A developer should see why a file is changing, which tool ran, what failed, what it cost, and what happens next. The loop is visible because invisible automation creates doubt.
NoteWhat it explainsThe agent that remembers.Why retrieval memory should be useful, visible, editable, exportable, and deletable.The shape of a good plan.How Plan mode turns uncertainty into a reviewable path before writes happen.Free, without the trap.How managed inference, sponsor disclosure, and user controls fit together.

Latest notes.

Why goals need contracts.

A long-running agent without a finish line is just a process. `/goal` starts with the objective, proof, budget, scope, and stop conditions. The contract is what lets autonomy stay useful instead of wandering.

Provider choice is product design.

A coding agent should not trap the work inside one model vendor. Picasso treats providers as routes through the same harness, so memory, tools, subagents, and goals travel with the work.

Subagents, without theater.

A subagent is not a mascot. It is a scoped worker with a job, a budget, and a result that returns to the main loop. Split the work. Keep the decision in one place.

Sponsor surfaces that stay in their lane.

Free software has a cost. Picasso names the sponsor, bounds the placement, and keeps sponsor systems away from code, prompts, memory, model output, tool output, and provider credentials.

Read next.

NoteWhat it explains
The agent that remembers.Why retrieval memory should be useful, visible, editable, exportable, and deletable.
The shape of a good plan.How Plan mode turns uncertainty into a reviewable path before writes happen.
Free, without the trap.How managed inference, sponsor disclosure, and user controls fit together.
Local when it matters.Why Ollama and vLLM belong beside frontier providers in the same product.

How we write.

Picasso writing follows the product. We name the behavior, show the boundary, and keep the builder's attention on the work.

Specific.

Notes point to real commands, modes, routes, controls, or product surfaces.

Disclosure first.

Sponsor-funded access is labeled clearly and kept separate from product claims.

No theater.

We would rather explain a limit plainly than hide it behind launch language.

The waitlist
Code Freely.

Picasso for Mac is almost here — a coding agent that looks the way serious tools should, and costs what creative freedom should: nothing. Leave your email and be first on the canvas.

Not live yet. Free for developers when it is — that's the point.

Sponsors and labs — the early canvas is yours. Choose Sponsor or Lab above and we'll reach out before launch.