Skip to content

v1.1.0 — Agents that build and train themselves

Choose a tag to compare

View all tags
@theguysudo theguysudo released this 06 Sep 12:39
· 21 commits to main since this release
v1.1.0
869ed9a

ENZO v1.1.0 — self-hosted, bring-your-own-key AI workspace.

Highlights

  • Custom Agent Builder — describe a task in plain English ("an agent that researches MUN country positions") and it builds itself. A two-pass drafter analyzes the domain first, then a free model from your own providers writes a 30-year-veteran operating manual — tacit knowledge, decision heuristics, edge cases — and picks the tools the task actually needs. You can also create agents straight from chat.
  • Self-improving agents (neural layer) — every agent keeps learning from your platform activity, whether you run it or not. A per-agent neural weight vector folds in domain-matched traffic on a 90-second cadence, distills expert lessons into memory on a schedule, and injects a live NEURAL FOCUS block into every run. Inspect it in the agent's Neural tab and train on demand.
  • Race drafting — when a draft needs a model, ~10 free candidates fire simultaneously and the first to answer wins. A brain-health scoreboard tracks which models actually deliver and reorders future races, so a dead or rate-limited free-tier model can no longer collapse a draft into a generic template.
  • First-key claim bootstrap — a fresh docker compose up now unlocks the entire platform from the web UI. Paste any working provider key; the first live-validated one claims the instance (written to the container .env and sealed AES-256-GCM into the enzo-memory volume) and every server-side feature — agents, skills, memory — unlocks immediately and survives restarts. No master key to configure. (Pre-seed via compose env if you prefer no claim window.)
  • Honest provenance — every agent records which model actually drafted it, and says so plainly when nothing was reachable.
  • ~20 hardening fixes across the terminal, agents UI and onboarding — research steps survive interrupted streams, tab switches no longer lose agent conversations, agent runs carry full conversation history (multi-turn confirm flows complete), reasoning models get proper token headroom, and the Google AI Studio key is fully optional in the self-hosted edition.

Upgrade

docker compose pull && docker compose up -d

Your data (projects, skills, memory, sealed vault state) lives in named volumes and survives the upgrade untouched.

Full changelog: docs/CHANGELOG.md

Apache-2.0. No accounts, no markup, no telemetry.

Assets 2
Loading