Skip to content

Axonlotl

Axonlotl is a self-healing infrastructure platform designed to solve the critical brittleness of multi-agent LLM systems, which currently face production failure rates of 41–86%.

Code Video Cerebral Valley

Core trick

Reroutes a failed agent's work to peers with overlapping capabilities and strengthens successful bypass routes with weighted updates.

Tags

agentsdev tools

Description

Axonlotl is a self-healing infrastructure platform designed to solve the critical brittleness of multi-agent LLM systems, which currently face production failure rates of 41–86%. While traditional software relies on deterministic fault tolerance (retries/catch blocks), Axonlotl implements a synthetic nervous system that treats agent networks as biological entities capable of autonomous repair.
preyashyadav/axonlotl★ 10 forksJavaScriptCreated May 23, 2026 · last push May 24, 2026

README excerpt

# Axonlotl

Brain-inspired self-healing infrastructure for multi-agent systems — built for the **Google I/O Hackathon 2026 (Cerebral Valley, May 23, 2026)**.

Axonlotl runs a multi-agent pipeline (planner → coder → reviewer → tester → deployer) and keeps it alive under failures, semantic drift, and latency by applying “brain protocols”:

- **Pattern 1 — Fast Path (Experience-Independent Baseline):** run normally when healthy
- **Pattern 3 — Lesion Recovery (Functional Rerouting):** bypass dead/critical agents using capability overlap
- **Pattern 2 — LTP (Experience-Dependent Structural Adaptation):** strengthen successful bypass “synapses” (route weights) over time

![Axonlotl cover](docs/screenshot/Gemini_Generated_Image_wkah96wkah96wkah.png)

![Axonlotl dashboard screenshot](docs/screenshot/image%20(2).png)

## What The Hackathon Demo Shows

- **Semantic drift detection:** middleware flags generic/low-quality outputs (e.g. “LGTM” on complex code)
- **Graceful degradation:** prompts get shorter / stricter when health is low
- **Compensatory rerouting:** when an agent is killed or drifts, Axonlotl reroutes to a backup agent in *coverage mode*
- **Neuroplasticity:** successful reroutes increase `routeWeights[A→B].weight`, making the same detour faster/stronger next time
- **“Managed Brain” analysis:** a managed Gemini agent (Axonlotl Core / “Prefrontal Cortex”) produces a 1‑sentence medical verdict after each run, with a safe fallback so the demo doesn’t break

## Dashboard “T

Team