OpenProse
An open-source declarative agent programming language written entirely in Markdown. It is event-driven rather than chat-based and is used to build multi-agent workflows; both the compiler and the virtual machine are themselves prompts.
Timeline (last 90 days)
OpenProse is an agent programming language written entirely in Markdown that is event-driven, declarative rather than procedural, open source, runnable locally via NPX, and also offered as a hosted version at run.prose.md.
OpenProse's compiler and virtual machine are themselves just huge prompts.
In a GitHub Webhook-driven PR review workflow demo, some PRs went through as many as 14 rounds of review feedback loops, and the system autonomously completed dozens of pull requests within a few days.
A large bank's quantitative models require updated regulatory documentation every three years (or upon changes); a single bank has about 4,000 models, and for larger institutions the related cost loss can reach $10 million to $50 million per year.
The team says several large companies have adopted OpenProse's permissively licensed open-source core to build their own stacks on top of their own harnesses and approved models.
OpenProse is the exact opposite of DSPy: DSPy treats every LLM call as a black box, with code and interfaces entirely deterministic and typed; OpenProse believes models are smart enough to be the compiler, putting everything in the fuzzy domain.
The team states clearly that OpenProse itself has no place in the Jet-class (one-shot classification) model domain, because it understands language rather than generates it; but it is particularly suited as a fast, cheap validator at the end of a workflow.
OpenProse hardcodes confirmation prompts for any operation that incurs cost.
Chat mode suits exploration mode (when you don't know what you want), whereas OpenProse targets scenarios requiring frequent execution of reliable, repeatable, auditable tasks.
Multiple contracts in OpenProse can be injected into the same run, and the agent must satisfy them all simultaneously; event-driven dependency propagation is similar to React's caching mechanism, where upstream state changes trigger downstream.
2 more timeline events
All Facts (12)
OpenProse is an agent programming language written entirely in Markdown that is event-driven, declarative rather than procedural, open source, runnable locally via NPX, and also offered as a hosted version at run.prose.md.
50%UnverifiedOpenProse's compiler and virtual machine are themselves just huge prompts.
50%UnverifiedIn a GitHub Webhook-driven PR review workflow demo, some PRs went through as many as 14 rounds of review feedback loops, and the system autonomously completed dozens of pull requests within a few days.
50%UnverifiedA large bank's quantitative models require updated regulatory documentation every three years (or upon changes); a single bank has about 4,000 models, and for larger institutions the related cost loss can reach $10 million to $50 million per year.
50%UnverifiedThe team says several large companies have adopted OpenProse's permissively licensed open-source core to build their own stacks on top of their own harnesses and approved models.
50%UnverifiedOpenProse is the exact opposite of DSPy: DSPy treats every LLM call as a black box, with code and interfaces entirely deterministic and typed; OpenProse believes models are smart enough to be the compiler, putting everything in the fuzzy domain.
50%UnverifiedThe team states clearly that OpenProse itself has no place in the Jet-class (one-shot classification) model domain, because it understands language rather than generates it; but it is particularly suited as a fast, cheap validator at the end of a workflow.
50%UnverifiedOpenProse hardcodes confirmation prompts for any operation that incurs cost.
50%UnverifiedChat mode suits exploration mode (when you don't know what you want), whereas OpenProse targets scenarios requiring frequent execution of reliable, repeatable, auditable tasks.
50%UnverifiedMultiple contracts in OpenProse can be injected into the same run, and the agent must satisfy them all simultaneously; event-driven dependency propagation is similar to React's caching mechanism, where upstream state changes trigger downstream.
50%UnverifiedThe team is building a router so that events are automatically routed to the corresponding task without having to manually click a button.
50%UnverifiedThe team says the combination of OpenProse with reinforcement learning has not yet been implemented, but they have discussed it with potential customers, because all contracts are text and thus naturally suited to version control and RL.
50%