​

Competitor Comparison & Migration Guide

Myrm is designed as a complete AI agent workspace. Unlike terminal-only coding agents, Myrm provides a GUI-first experience with persistent sandboxes, cross-session memory, and enterprise-grade security — all while maintaining full coding capabilities.

​

What you gain by migrating

Most migrations are not “move a chat history”. They are a chance to replace five disconnected tools with one workspace that has a memory, a control surface, and an exit path. The honest part: there is no one-click import from closed Personal Agents, because their memory and bot graph have no public export format. Migration means keeping the portable assets and rebuilding the private ones. We say that before you commit, not after.

​

At a Glance

Cells marked — are outside the verified scope, not a claim that the competitor lacks the capability. The knowledge-base row reads each competitor’s source as of 2026-11-01.

​

Muse and closed Personal Agents: what shipped, what you can own

Muse’s November 2026 changelog is a useful public checklist: 18 capabilities since launch. It is also a fair reminder of where closed consumer agents are strong. Below is the honest hand-off — “roadmap” means not shipped, and we do not count it as supported.

​

What you keep, what you rebuild

There is no one-click import from Muse because its private memory and bot graph have no public export format. Any “100% seamless” claim would be false. What you gain is ownership: MIT product layers, self-hosting, BYOK, 26 channels, white-box memory, deterministic approvals, diff/rollback, and the same data model on local, desktop, and cloud.

​

Where Muse is still ahead (honest)

• Native consumer apps and distribution: App Store / Play Store presence and native share sheets are ahead of our PWA route.
• Consumer emotional design: plush characters, naming rituals, and a calm consumer tone are more mature; our companion covers part of this, not all of it.
• Global regional coverage: Canada, Japan, and Singapore are live for Muse; we do not claim equivalent coverage.
• Hardware gadget SDK and outbound calling: Muse ships a gadget line and an outbound-calling beta; both are roadmap items for us.

The trade: if you want the most polished consumer app today, Muse is a strong choice. If you want an agent workspace you can own, audit, extend, self-host, and take with you — including your keys, skills, MCP servers, rules, and exported memory — Myrm is built for that. We name the gaps instead of hiding them.

​

Usage policy: no bait-and-switch

A recurring complaint about subsidized consumer agents is the cycle: heavy usage at launch, users build habits, then limits are cut or prices rise. The pattern is hard to verify from outside, but the trust damage is real. Myrm is structured differently:

• BYOK means your provider bills you directly. The platform does not subsidize tokens and then recover the cost by cutting you off. You can see the provider’s price and change models at any time.
• Platform plans show budgets and limits up front. Daily/session caps, four-tier spend control, per-agent usage share, and a burn-down forecast are visible before you hit a wall, not after.
• Power users are not a single bucket. BYOK, local models, platform plans, and per-task routing let a heavy user pay for what they use instead of forcing every user into one quota.
• No undated promises. We do not market “unlimited forever” without a date or a boundary. If a platform plan changes, the honest commitment is notice plus an exit: keep your keys, skills, MCP servers, prompts, rules, and exported memory, and continue with BYOK or local models.

Honest boundary: platform WU plans can change; we cannot promise “never change”. What we can promise is that the platform does not depend on “give it away, then nerf it” as a growth model, and that BYOK is always the exit.

​

Team Bot, Data Bot, and why Slack matters

Grok Bot’s November 2026 Team Bot post is a useful adoption signal: a company-wide Data Bot handled an estimated 70 hours of data-scientist work in a week, and its own analysis found that 60% of messages for “broad and sticky” bots came from Slack, versus 12% for low-retention bots. The lesson is not “build a data bot”; it is “put the agent where people already work”.

• Channel-native, not another app: Myrm ships 26 channel providers (Slack, Feishu, WeChat, Telegram, Discord, email, and more) with the same memory, approvals, artifacts, and task state. A Slack thread can start a task; the result can come back to the same thread.
• Data work is a template, not a separate product: code execution, datasets, spreadsheets, interactive canvas, and artifacts cover the data-analyst workflow; 22 role templates include financial and research analysts. A dedicated “Data Bot” template is content work, not a missing core capability.
• Shared team bots are a deployment mode: Myrm’s model is one Agent cloud computer per user. Cross-employee shared agents belong to the enterprise multi-tenant publishing path (permissioned app, read-only shared knowledge, per-member memory isolation); we do not claim a “one shared bot computer for the whole company” today.
• Usage metrics stay aggregate: token, cost, and call counts are aggregated per agent; prompt content is not part of the metrics. The Data Flow & Privacy Disclosure panel shows what egresses.

Honest boundary: if your requirement is literally “everyone in the company uses one shared bot on one computer”, we do not ship that; our isolation model is per user, and shared publishing is on the roadmap. If your requirement is “the agent lives in Slack/Feishu/WeChat and keeps one memory and one approval surface”, Myrm’s channel-native design is broader than a Slack-only Team Bot.

​

Reasoning: visible, stoppable, replayable

Most agents leave a static skeleton while the model thinks. Myrm makes the chain part of the message:

• Live: reasoning deltas stream into the assistant message with a real-time timer.
• Stoppable: “Stop thinking” ends reasoning early and answers with what is known; bounded to 10 s so the button cannot get stuck.
• Collapses once (1 s after the answer starts) and stays reopenable.
• Real tokens: provider-reported reasoning_tokens first, an explicit ≈ estimate only as a fallback.
• Three persisted modes: off / collapsed / inline; off stops the server from sending reasoning at all.
• 6 languages + 40 px touch targets, fully localized copy and ARIA labels.

​

Model routing & provenance

Most agents pick a model for you but do not tell you which one answered. Myrm exposes the routing decision per answer:

• Two routing engines in production: task-complexity routing (SIMPLE / STANDARD / REASONING) and task-specialty routing (CODE / LONG_DOC / REASONING / MULTIMODAL / CASUAL / GENERAL), each with a fallback chain.
• Per-answer provenance: the usage popover shows the actual model, routing tier, task specialty, routing reason, and cost.
• Opt-in per Agent: pick light/reasoning models under Smart Routing (3 Pareto presets ship configured). Not default-on for every Agent.
• Deterministic approvals: whether an action needs your approval is computed from rules, not by a classifier. Spend and irreversible actions have no “always allow” button; a secret typed into the masked card is bound to one command and consumed once.

​

Hands-off delegation

The most common complaint about always-on agents is babysitting. Myrm is built to be left alone:

• Durable background work: offline tasks survive disconnect/restart and are resumed from the database; the desktop keeps the machine awake.
• Completion reaches you: 26 channel providers, in-app notifications, and a Catchup inbox.
• No empty pings: cron jobs support a [SILENT] contract.

​

Skill quality

Installing a skill adds text to the context — it does not change the model. The real question is: did the agent follow the skill, and did that improve the result?

• Shipped: three-state attribution (selected / used / observable), 26-category / 113-pattern static scanning, permission templates with a skill doctor, progressive disclosure.
• Project-scoped skills: any SKILL.md under the workspace (depth 3) is discovered behind a trust gate. .claude/skills/ works as-is.
• Anthropic’s public skills work as-is: name / description / license / compatibility / metadata / allowed-tools are parsed; versioning, rollback, A/B evaluation and permission tightening are added on top.

​

Agent teams: verification over headcount

Competitors market agent counts. Agent count is not capability — a team that cannot verify its work is just noise.

• Shipped: 5 team templates, Kanban task DAG, goals acceptance criteria with shell/semantic checks, completion delivery over 26 channels, fleet KPI board.
• Unattended approvals are bounded to a declared capability set.

​

Complex work: completion, verification, handoff

​

Memory backends: Myrm vs hosted memory services (GBrain)

GBrain (Oct 2026) sells a hosted Postgres + pgvector memory that any agent can point at. The pitch is real — setup friction is the top reason non-technical users quit — but it concentrates every agent on one shared vector store with application-level scoping, which is exactly what its own comment thread worries about.

• Isolation by construction: each Myrm user’s memory lives in their own sandbox (own SQLite + own vector index). There is no shared store to scope.
• Write governance: contradictory writes land in a review queue; high-risk contradictions never auto-resolve; deletes move to an archive with explicit purge.
• Exit cost: the vector index is rebuildable (estimate → run → status) and memory exports carry SHA-256 checksums.
• Honest gap: “paste one link and the agent configures itself” is smoother than our copy-a-snippet flow. We ship a portable Agent Plugins bundle and step-by-step snippets instead of a fetchable manifest.

​

Updates and recovery: Myrm vs OpenClaw v2026.10.9

OpenClaw’s patch release is entirely about update failures: rollbacks that preserve local edits, recovery from stale indexes, and restarts that no longer leave the gateway stopped. Credit where due — that is the right priority. We are candid about where we stand: our data layer is stronger (page-level salvage, checksummed backups, reversible state swaps), while our source/CLI updater is still a plain git pull with no staged validation or rollback. Staged upgrades, a cross-version regression gate and a rollback entry point are on the roadmap (#167 / #215 / #478); we do not claim them until they ship.

​

Personal Agent era: Myrm vs Muse / Today / GrokBot / Instinct / Dots

Closed Personal Agents (Meta Muse, Today, xAI GrokBot, Instinct, OpenAI Dots) are optimized for emotional polish and zero-config onboarding. Myrm is optimized for ownership, control, and extensibility. We are explicit about the trade-off: if you want the most polished consumer companion today, those products are ahead on 3D avatar, calm replies, message reactions, naming rituals, and single-assistant presentation. Those details are on our roadmap (#496 / #497 / #498 / #89) — we do not claim them as shipped.

​

What actually carries over

There is no one-click import from a closed Personal Agent: their memory and bot graph have no public export format. Migrating means rebuilding those parts while keeping your keys, skills, MCP, prompts, and channels. We do not claim otherwise.

​

Claude Code’s 7-layer stack: what carries over

The widely shared “7 repos that upgrade Claude Code” list is a stack, not a product. Layer by layer, here is what a Claude Code user keeps when moving to Myrm — and what they should not expect. Migration: .claude/skills/, SKILL.md, plugin bundles, MCP configs, BYOK keys, and project rules carry over. What changes is the surrounding workspace: long-term memory, 26 channels, Kanban, approvals, and one server for local, desktop, and cloud.

​

Where competitors are still ahead

A comparison page that never says “they are better here” is a brochure, not a comparison. As of 2026-11, these are the areas where we do not claim to lead: How to read the rest of this page: ✅ means we found the capability in the product or source we inspected; — means it was outside the verified scope, not proof that the competitor lacks it. We update the “as of” dates when we re-check.

​

Migration path

Myrm runs as SaaS, self-hosted WebUI, or desktop. Your skills and memory follow via the GUI migration wizard (desktop / local WebUI).