Product / UX / Orchestration & agents

Vellum

Consumer-facing 'Personal Intelligence' agent, a multi-platform (web, macOS, iOS, CLI) assistant with episodic, semantic, and procedural memory that takes actions on the user's behalf. Pivoted in 2026 from an LLMops/prompt-engineering platform (prompts, workflows, evals) to a consumer agent product after a $20M Series A in July 2025; legacy LLMops surface still available but no longer the headline.

Vellum (vellum.ai), a PERSONAL AI ASSISTANT/agent ('the AI layer for your life', 8-type structured memory, multi-channel autonomous action across Mac/Web/iOS/Slack/Telegram, proactive hourly check-ins, out-of-process credential isolation). The agent's full source ships MIT-licensed at github.com/vellum-ai/vellum-assistant (backend gateway, multi-platform clients, skill framework); a managed cloud (memory/security) is offered on top. Verified live June 2026; positions itself vs Hermes Agent / OpenClaw.

Openness

5 high confidence

5.0

license: MIT(OSI, Copyright 2025 Vellum AI)
source: public(the full agent - gateway, clients, skill framework, memory - in github.com/vellum-ai/vellum-assistant)
core-gated: ungated(README: managed runtime on Vellum Platform or self-hosted, "Same codebase, same data model")
self-host: yes(by design)
managed-cloud: hosting convenience, not a feature tier

CORRECTED from the record's "closed/score 1". The agent's complete source is public and MIT-licensed (LICENSE: "Copyright (c) 2025 Vellum AI"), covering the backend gateway, the Mac/Slack/Telegram clients, the sandboxed skill framework and the memory system, and it is self-hostable by design rather than a client stub. The vendor's own marketing comparison table labels Vellum "Proprietary" against MIT for Hermes/OpenClaw, because it sells a managed cloud tier. Score corrected from 4 to 5 on 2026-07-30. The earlier 4 rested on the managed memory/security layer being a differentiator, but the README states the managed runtime and the self-hosted one are the "Same codebase, same data model", so hosting is convenience rather than a feature tier. The ladder's core_gated note is explicit that a hosted offering alongside an ungated core is not gating, and 4 pairs with open_core in any case.

https://www.vellum.ai recorded 2026-06-04
personal AI assistant 'AI layer for your life'; comparison table labels Vellum 'Proprietary' vs MIT Hermes/OpenClaw, but links to its own GitHub; cloud or self-hosted
https://www.vellum.ai/blog/best-hermes-agent-alternatives recorded 2026-06-04
Vellum positioning as a personal assistant vs Hermes/OpenClaw (confirms entity)
https://github.com/vellum-ai/vellum-assistant recorded 2026-06-04
PRIMARY: full open source agent source (gateway+clients+skill framework+memory), MIT, ~577 stars, 120+ releases (latest v0.8.7 Jun 2026)
https://github.com/vellum-ai/vellum-assistant/blob/main/LICENSE recorded 2026-06-04
PRIMARY: MIT License text, 'Copyright (c) 2025 Vellum AI'

Adoption

1 low confidence

1.0

No download/user/usage metrics disclosed (it's a self-hosted app, not a registry package). Only signal available is GitHub stars (~577 on vellum-ai/vellum-assistant), used here strictly as a last-resort proxy per the adoption!=stars rule; small/new entrant. Previously null; set to a conservative level 1 now that a real (if weak) signal exists.

https://github.com/vellum-ai/vellum-assistant recorded 2026-06-04
~577 stars (last-resort star proxy; no other adoption signal disclosed)
https://www.vellum.ai recorded 2026-06-04
landing page with no disclosed user/usage numbers

Capability

3 low confidence

3.0

Solid personal-agent feature breadth (8-type memory + multi-channel autonomous action + credential isolation + proactive engine), now confirmed against the open MIT source. No benchmark and no frontier differentiator; scored 3 on feature matrix.

https://www.vellum.ai recorded 2026-06-04
8-type memory, multi-channel autonomous action, credential isolation, ambient engine features

Unchanged since 2026-07-30 (last edited, not re-checked)