Clojurians Log v2

⚒️ dvergr — an FRP programming model for agents, in a discourse framework Agents as reactive processes in continuous-time rooms; humans, LLMs, and scripts are participants in the same shape, composing through tagged messages on a small pub/sub kernel. You program agents and compose them into multi-agent workflows — and because they run in the same live medium they're built from, an agent can even fork a world, wire up a workflow, and hand it back for you to review (early + exploratory, not the default). Built on the replikativ stack: • Spindel — functional reactive runtime (CoW context forking + pub/sub) • Datahike — immutable Datalog (conversation + knowledge graph) • Yggdrasil — copy-on-write branching across git + database Batteries included: a safe SCI clojure_eval sandbox with a real agent-owned git workspace + jailed shell, ~25 read-only intake sources, boundary secret injection (agents use API keys they never see), pluggable LLM providers (Anthropic / OpenAI / Fireworks / local claude CLI, zero-key), per-turn microdollar budgets, context compaction, and one daemon driving TUI / web / Telegram / nREPL. Apache-2.0 · <https://github.com/replikativ/dvergr> Live notebooks: <https://replikativ.github.io/dvergr/> Early days — feedback, issues, and clojure_eval war stories all very welcome. clojure