Skip to content

Roadmap

What we're building next, what comes after, and what we set aside. Each line links to its issue on GitHub, where the work happens in the open, or to the episode where we promised it. What works today is on What works today.

Next

Chat with one general agent

Coding with Coder

Across your devices

Models and the OpenAgents API

Gym, evals, and proof

Traces and your data

  • A plain-words changelog links each change to the work behind it. Web. Follow it: #11104. Promised in Episode 272.

Agent-ready and open

Surfaces

Later

Account and identity

Chat with one general agent

Coding with Coder

Across your devices

  • You get a phone notification when an agent needs you. Phone. Promised in Episode 160, Episode 183, Episode 190.
  • Your progress, levels, and rewards are the same in every app. Web, Terminal, Phone, Desktop. Promised in Episode 284.

Models and the OpenAgents API

  • You say what you care about (cost, speed, privacy) and the agent picks models to match. Web, Terminal, Phone, Desktop. Follow it: #11079. Promised in Episode 269, Episode 286.
  • A public counter shows tokens served and which models served them. Web. Follow it: #11081. Promised in Episode 243, Episode 244.
  • The app never claims a model it didn't use; you get the model you picked or a visible failure. Web, Terminal, Phone, Desktop. Promised in Episode 250.
  • You can run open models on your own machine or a friend's. Terminal, Desktop. Follow it: #11080. Promised in Episode 145, Episode 194, Episode 219.
  • You can pay more for private processing of sensitive work. Open network. Promised in Episode 242.

Plugins and new abilities

Gym, evals, and proof

  • Gym measures agents and plugins, and anyone can inspect the results. Web, Terminal. Promised in Episode 65, Episode 287, Episode 289.
  • We publish benchmark results with the setup, costs, and failures, not only scores. Web. Promised in Episode 120, Episode 121, Episode 186, Episode 287.
  • You earn experience, levels, and leaderboard places for verified work, never for wasted spending. Web, Terminal, Phone, Desktop, Verse. Promised in Episode 185, Episode 270, Episode 284.
  • Leveling up unlocks free credits, better models, and hosting. Web, Terminal, Phone, Desktop. Promised in Episode 284.
  • Improvements to how the agent works are published with their evidence, so everyone's agent gets better. Open network. Promised in Episode 270, Episode 286, Episode 287.
  • You can report something that doesn't work, and it becomes a tracked fix. Phone, Web, Terminal. Promised in Episode 234, Episode 246, Episode 254.
  • We publish accepted outcomes per dollar, and later per kilowatt-hour, as our main measure of work. Web. Posted on X: 2026-06-08, 2026-07-20.

Traces and your data

Wallet and payments

Markets, compute, and earning

Agent-ready and open

Surfaces

The Verse

Dropped on purpose

Earlier ideas

  • A drag-and-drop visual builder for agent workflows. Set aside for chat (076); the agent now writes and tests the ability itself (F1). Web, Terminal, Phone, Desktop. Promised in Episode 38, Episode 58, Episode 59, Episode 61, Episode 73.
  • A store of many separate agents to browse and rate. Replaced by one conversation backed by a network of agents and plugins (086, 289). Web, Terminal, Phone, Desktop. Promised in Episode 90, Episode 92, Episode 94, Episode 100.
  • An open replacement for WordPress, and sites built by asking. OpenPress and Sites were retired. A site is a coding task for Coder. Web, Terminal, Phone, Desktop. Promised in Episode 126, Episode 127, Episode 129, Episode 180, Episode 229.
  • A built-in CRM and an all-in-one business suite. Outside the current product focus; Coder and the general agent come first. Web, Terminal, Phone, Desktop. Promised in Episode 135, Episode 237, Episode 239, Episode 247.
  • Code hosting to replace GitHub. "Too big an apple" (281). Back to GitHub. Web, Terminal, Phone, Desktop. Promised in Episode 270, Episode 272, Episode 273, Episode 274.
  • Paid distributed training on home machines. Paused; not part of the current product. Our inference engine stays for local use and research. Web, Terminal, Phone, Desktop. Promised in Episode 216, Episode 238.
  • Liquidity and risk markets. Speculative, with no buyers. Web, Terminal, Phone, Desktop. Promised in Episode 213, Episode 214, Episode 230.
  • Ocean-powered data centers. Out of scope. Web, Terminal, Phone, Desktop. Promised in Episode 194, Episode 227.
  • One-click security scans of open-source projects. Stood down in 266. Web, Terminal, Phone, Desktop. Promised in Episode 264, Episode 265.
  • A named voice agent that runs things for you (Onyx, Sarah). Replaced by the one OpenAgents agent (289). Voice returns as B12. Web, Terminal, Phone, Desktop. Promised in Episode 139, Episode 151, Episode 260, Episode 270.
  • Pay-per-message pricing in sats. Replaced by a free tier, Pro, and credits. Web, Terminal, Phone, Desktop. Promised in Episode 97, Episode 118.
  • Paying providers per second online. Dropped in 224 for paying for accepted work. Web, Terminal, Phone, Desktop. Promised in Episode 214, Episode 223.
  • Hand-gesture "Jarvis" control. Teaser only. Web, Terminal, Phone, Desktop. Promised in Episode 176, Episode 179, Episode 183.
  • Desktop with no OpenAgents account. Replaced by one account everywhere (A1). Local-only use of Coder stays. Web, Terminal, Phone, Desktop. Promised in Episode 248, Episode 251.
  • Small models running on the phone. Too slow on Android (149). Web, Terminal, Phone, Desktop. Promised in Episode 145, Episode 149.
  • Public investigation agents (Sleuth, OSINT) and public knowledge graph. Domain experiments with no follow-up. Web, Terminal, Phone, Desktop. Promised in Episode 43, Episode 44, Episode 146, Episode 154.

Want something that isn't here? Ask for it on GitHub.