Skip to main content

Compare

Bernstein vs Muse: quick decision guide

Muse Code CLI adapter for Bernstein.

Page built on 2026-08-10 from data/adapters-meta.json. Every claim below links to its primary source.

Install both

Muse

curl -fsSL https://dev.meta.ai/install.sh | sh

Bernstein

pipx install bernstein

Apache-2.0. Deterministic Python scheduler.

Feature matrix

CapabilityMuseBernstein
Install methodcurl -fsSL https://dev.meta.ai/install.sh | shpipx install bernstein
LicenseNot recordedApache-2.0
AuthenticationNot recordedPer-agent credential scoping (no shared key)
Multi-agent orchestrationOne agent in a terminalMuse plus 40+ other adapters in parallel worktrees
MCP supportNot measuredYes
Parallel-safe in worktreesNot measuredYes (designed around git worktrees)
HMAC-chained audit logNoYes (RFC 2104 SHA-256 chain in .sdd/)
Deterministic schedulerNot applicable (single-agent CLI)Yes (Deterministic Python scheduler)

Adapter source: src/bernstein/adapters/muse.py | Upstream homepage: dev.meta.ai

Verifiable facts

The brief for this surface requires at least three facts that a reader can verify against a primary source. The list below is built from the bernstein adapter source and, when available, the upstream project's own pages.

  1. Bernstein ships a Muse adapter at src/bernstein/adapters/muse.py that wraps the upstream CLI as one of 42 routable agents. [source: bernstein adapter source, as of 2026-08-10]
  2. Upstream install command, as recorded in the bernstein adapter, is "curl -fsSL https://dev.meta.ai/install.sh | sh". [source: upstream docs, as of 2026-08-10]
  3. Bernstein is an open-source Multi-agent orchestrator licensed Apache-2.0, with a deterministic Python scheduler that routes work across CLI agents in parallel git worktrees. [source: bernstein repo, as of 2026-08-10]

Where Muse fits in Bernstein

Bernstein registers Muse under the slug "muse" and the registry name "muse". The adapter source lives at src/bernstein/adapters/muse.py in the bernstein repo and was last touched at build time 2026-08-10. The Muse adapter file is 253 lines and 9,958 bytes long, fingerprinted d1c0387f9cb9f789 (first 16 hex chars of SHA-256). Operators install Muse on a worker box with "curl -fsSL https://dev.meta.ai/install.sh | sh" before Bernstein routes any task to it. No upstream GitHub repository is recorded in the bernstein adapter for Muse; refer to the upstream vendor's documentation when auditing. The Muse project's homepage at dev.meta.ai is the primary source for upstream release notes. The bernstein adapter file for Muse does not yet carry a "Last verified against upstream" line; this means the adapter still tracks an unpinned upstream binary. Bernstein routes tasks to Muse when its pass rate on similar work clears the configured threshold, otherwise the deterministic Python scheduler picks a different adapter from the 40+ adapter catalog.

Adapter source excerpt

The module docstring of the Muse adapter, as it stands in the bernstein repo. Punctuation is normalised for the web; the wording is the author's. Length: 1549 characters.

Muse Code CLI adapter for Bernstein. Adapter for Muse Code, Meta's terminal coding agent (https://dev.meta.ai/docs/muse-code). Bernstein drives its documented headless mode - ``muse exec "<prompt>"`` runs a single prompt to completion, intended for scripts and CI - so a Muse Code run gets the same worktree isolation, journaling, and receipts as every other adapter-managed worker. Last verified against vendor docs on 2026-08-10. Install: ``curl -fsSL https://dev.meta.ai/install.sh | sh`` (static binary, macOS/Linux only; no native Windows support). Auth: ``META_API_KEY`` env var for non-interactive environments. Default model: ``muse-spark-1.2``. Version probe: ``muse --version``. Flags used (verified from the vendor configuration/permissions pages): * ``--model <id>`` - common launch flag, passed before the subcommand. * ``--disable-approval`` - skips approval prompts while keeping the CLI's own sandbox containment; required for unattended runs (the interactive approval prompt would hang a headless worker forever). ``--yolo`` also exists but additionally disables the vendor sandbox, which this adapter deliberately does not do. TODO-verify (documented but not consumed yet - re-check at https://dev.meta.ai/docs/muse-code/extending.md before wiring up): * ``--json`` (headless-only JSONL event stream) - unused; the adapter reads plain text output through the standard text-signal channel. * ``--session-id <uuid>`` resumes an existing session non-interactively; resume stays declared unsupported until the resume path supplies one.

Adapter telemetry

Registry namemuse
Adapter classMuse
Source filesrc/bernstein/adapters/muse.py
Source file size253 lines, 9,958 bytes
Source SHA-256d1c0387f9cb9f789d668b445f3fe9bef4b75890c1c4f25d77bbc6d8dc0333841
Category bucketcloud-sdk-family
Upstream repoNot derivable from adapter source
Upstream homepagedev.meta.ai
Last verified upstreamNo "Last verified" line in adapter source
Operator-curated overlayNo (programmatic page)

When to pick which

Choose Muse

Reach for Muse when the work is a single thread that fits one agent: in a single-process terminal session, designed for single-instance use per repo. Auth model is configured per upstream docs. You skip the orchestrator round-trip and get the smallest possible surface between you and the model.

Choose Bernstein

Wrap Muse under Bernstein when the goal splits into parallel tasks, when you want an HMAC-chained audit log on every routing decision, or when a deterministic Python scheduler (no LLM picking who runs what) is a hard requirement. Bernstein wires the cloud SDK adapter to the same audit chain the local CLI adapters use, so one regulator review covers both.

FAQ

Does Bernstein replace Muse?

No. Bernstein wraps Muse as one of 40+ CLI adapters and routes tasks to it based on per-task pass-rate history. Muse keeps running unchanged; Bernstein decides when it gets work.

Can I run Muse alongside other agents in the same repo?

Yes. Each agent runs in its own git worktree under .sdd/worktrees/{session_id}, so file edits never collide. Bernstein merges results back to the trunk only after the configured quality gates (lint, types, tests) pass.

Is this comparison page handwritten?

No. The template is fixed; every fact and every link is pulled from the bernstein adapter source in the master branch and (when available) the upstream project's own pages. The data extractor lives at scripts/gen-compare-data.mjs. No LLM writes the prose.