Ecosystem Map
A curated directory of the layer above frontier models — the frameworks, orchestration, marketplaces, and providers ClawBlog covers. Interactive directory of tools, services, integrations.
Curated snapshot — figures as of mid-2026.
The open-source runtimes that turn a frontier model into an agent that acts in the world.
The open-source AI-agent framework that defined the category — operated through messaging platforms (Discord, WhatsApp, Telegram) with 100+ AgentSkills. v2026415 ships native Claude Opus 4.7 integration; a non-profit foundation now stewards the project.
347K+ stars — most-starred repo in GitHub history
Nous Research’s self-improving agent with persistent memory. MIT-licensed, runs on five backends (local, Docker, SSH, Singularity, Modal), and was the fastest-growing agent framework of 2026 by early-adoption metrics.
95K+ stars — MIT-licensed
The layer above a single agent — coordinating many agents, enforcing budgets, keeping the lights on.
Run-it-for-you agent infrastructure — the hosted alternative to standing up your own runtime.
Where agent capabilities are published, discovered, and — as the ClawHavoc incident showed — attacked.
The frontier labs whose models sit underneath every agent in this directory.
Maker of the Claude model family (Opus 4.7, Sonnet 4.6, Haiku 4.5) and Claude Managed Agents — the models most agent frameworks in this directory reach for first.
Claude Opus 4.7 · Sonnet 4.6
Maker of the GPT-5.4 model family, including the GPT-5.4 Image 2 generation model widely used across the agent tooling ecosystem.
GPT-5.4 model family
Maker of the Gemini family and the Nano Banana image models — including the high-fidelity Nano Banana Pro released April 2026.
Gemini 3 · Nano Banana Pro
Mastra's latest release restores agent state without re-reading the whole conversation. The fix exposes a cost problem most agent users never knew they were paying: every resumed thread re-bills the entire history.
When Python packages can be published straight to the browser, the agent runtime stops being a server problem. Two shipping releases this week say the harness is moving client-side.
Apple licensed a Gemini-derived model and pointed vision LLMs at the screen instead of building an integration layer. That single choice sidesteps the harness problem every rival agent has been paying down by hand.
Browser-Use's 0.13.0 rewrite ditched the browser abstractions everyone assumed agents needed. Read against Apple's Siri, Claude Code's safe-mode, and the new code-quality benchmarks, it's a signal about where agent value is migrating.
Satya Nadella has shifted into a hands-on operating role at Microsoft. Read alongside the rest of this week's quiet release notes, it signals where the agent and AI infrastructure stack is consolidating, and which layer your agents will run on.
Mastra's new notification-inbox system lets agents send you persistent, priority-ranked messages that survive across sessions. The framing is mundane; the implication is that agents are quietly becoming collaborators you check on, not tools you run.
Agents are graduating from API calls to direct computer control. A new infrastructure layer is forming underneath them, and it's quietly rewriting what the word 'agent' means.
Arize Phoenix v16.0.0 ships Code Evaluators that let users write their own scoring logic in the UI, no deployment required. The real story is what this admits about the state of agent evaluation.
Mastra's new fine-grained access control and favorites system signals that agent frameworks are moving beyond single-user experimentation into multi-tenant governance.
