Hermes Agent adds MiniMax M2.7 and MiMo V2 Pro through partner integrations

• Nous Research said Hermes Agent now has a product and model partnership with MiniMax, and MiniMax confirmed that MiniMax M2.7 is already live inside Hermes through the Nous Portal.
• Nous Research also added Xiaomi's MiMo V2 Pro to Hermes through the same portal flow, with free access for the next two weeks after a hermes update.
• Supermemory launched a Hermes plugin promising shared memory across Hermes, OpenClaw, and Claude, and Teknium said it is now an officially supported memory plugin.
• Nous Research showed the Manim Community Edition core team contributing more API coverage to Hermes' Manim skill, which already runs a full animation pipeline according to the skill README.
• Hermes shipped the plumbing for this a few days earlier: the v0.7.0 release notes added pluggable memory providers, while the Quickstart documents provider switching through hermes model.

You can browse the Quickstart for the provider flow, read the memory provider plugin docs for the new extension point, and inspect the Nous Portal page, which describes model access, API keys, and usage monitoring. The Manim skill README says Hermes handles planning, code generation, rendering, scene stitching, and iterative refinement, while Nous Research's screenshot shows the CE maintainers filling in missing objects, animations, and equation environments.

The immediate product change is more hosted model choice inside Hermes. Nous Research framed MiniMax as a broader partnership, and MiniMax said M2.7 is already powering Hermes Agent through the Nous Portal.

That fits Hermes' existing provider model. The Quickstart says hermes model lets users switch inference providers, and the Nous Portal page positions the portal as Nous' managed path for model access, API keys, and usage monitoring.

MiMo V2 Pro looks more promotional than infrastructural, but still notable. Nous Research offered Xiaomi's model free for two weeks and tied access to the same hermes update path, which makes the portal look increasingly like Hermes' hosted model marketplace.

This is the cleanest example of Hermes' new plugin surface turning into an ecosystem. Supermemory pitched hybrid memory, automatically refreshed profiles, scoped memories for different tasks, and one shared memory layer across Hermes, OpenClaw, Claude, and other agents.

The official docs make that possible in a straightforward way. The memory provider plugin guide says providers live under plugins/memory/, implement the MemoryProvider interface, and extend Hermes beyond the built in MEMORY.md and USER.md. The v0.7.0 release notes described pluggable memory providers as one of the headline changes, and Teknium said SuperMemory is now an officially supported plugin rather than a one-off hack.

The Manim push is less about a brand partnership and more about making one of Hermes' flashiest skills less brittle. Nous Research's screenshot says Manim Community Edition contributors added coverage for:

• new mobjects including Polygon, Arc, Sector, Annulus, and matrix families
• animation methods including GrowFromPoint, GrowFromEdge, SpinInFromNothing, MoveAlongPath, Rotate, and Rotating
• equation support for bmatrix, pmatrix, vmatrix, cases, aligned, and tex_environment
• a version pin to Manim CE v0.20.1 in SKILL.md

The Manim skill README says Hermes already uses the skill for 3Blue1Brown-style technical videos, handling planning, Python generation, rendering, scene stitching, and iterative refinement. The extra API coverage matters because it expands the documented surface the agent can reliably call when it builds those animations.

Two smaller tweets filled in behavior that the main announcements skipped. In a reply about setup friction, Nous Research suggested hermes claw migrate, a command meant to help OpenClaw users move into Hermes.

And in the Manim thread, Nous Research said the skill uses ElevenLabs for voiceover. That is a concrete implementation detail missing from the launch clip, and it rounds out the picture of Hermes as a skill hub that is adding hosted models, external memory backends, migration paths, and media tooling at the same time.