awesome-agent-evolution
A community-curated “awesome list” from evomap (evomap.ai) cataloging 150+ projects, papers, and tools on AI agents that improve themselves — “self-evolution, memory systems, autonomous self-improvement, and the infrastructure that powers them.” Where this spoke has built its map of the field bottom-up from individual tools, this is a top-down taxonomy of the whole area by an outside curator, so its main use here is as a field map to check the spoke’s layer model against and to surface named projects worth their own pages later.
The taxonomy
The list is organized into twelve sections, which line up closely with layers the spoke arrived at independently:
- Agent Evolution and Self-Improvement — the spoke’s self-improving-agents growth axis.
- Memory Systems — agent-memory (the list names Mem0, ~60k★, as a canonical product, alongside the spoke’s seekdb/memory-vault/recall stratum).
- Agent-to-Agent Protocols — the interop edge, a2a-protocol.
- Agent Development Platforms — the agentic-coding-harness / vendor-platform stratum (langchain and others).
- Agent Coding and Software Engineering — coding harnesses (Claude Code is listed at ~135k★).
- Prompt and Behaviour Optimization — the authoring-shift toward loop-engineering.
- Agent Safety and Guardrails — agent-guardrails.
- Embodied AI — a corner the spoke has not touched (robotics/physical agents).
- Key Research Papers — surveys and primary literature (the spoke’s self-evolving-agents-survey sits in this tradition).
- Benchmarks and Evaluation — a section pointing at where field-level evaluation lives.
- Taxonomy and Community and Knowledge — the framing/meta sections.
Notable named entries
As a landscape marker (not yet paged here): Eliza (~18.6k★, autonomous agent framework), Agent Zero (~18.3k★, “learns through interaction”), and Mem0 (~60k★, production agent memory). These are page candidates if a source arrives on any of them directly — the list gives a one-line description and a star count, not enough for a grounded page on its own.
Why it matters here
Two things. First, it is external corroboration of the spoke’s structure: an unaffiliated curator carves the field into self-improvement / memory / interop / platforms / guardrails / benchmarks — nearly the same divisions the spoke reached from the tools up, which is mild evidence the layer model is the field’s shape, not just this wiki’s framing. Second, its Benchmarks and Evaluation section is a lead on the spoke’s standing #1 open question (neutral, measured comparisons) — but a directory of benchmarks is a pointer, not a benchmark; it says where to look, it doesn’t settle “does structure substitute for capability, and how far.”
Provenance
- Maintainer: evomap (evomap.ai), which publishes the list and an accompanying visual taxonomy.
- Tier: T3 — a self-curated community aggregator (selection and star counts are the curator’s; entries are not independently evaluated). Star counts and the 150+ figure are volatile — snapshot 2026-07-02.
Related
self-improving-agents · agent-memory · a2a-protocol · agent-guardrails · agentic-coding-harness · self-evolving-agents-survey · loop-engineering · synthesis