Skip to main content

isomer-labs

Documentation: https://codegandee.github.io/isomer-labs/

Isomer Labs is a research workbench for running multi-agent research topics from a project directory. It gives operators a CLI, packaged system skills, topic workspaces, durable records, and a local web GUI for inspecting how ideas, experiments, and artifacts evolve.

Note: Isomer Labs is designed to be driven by a CLI agent. The recommended usage pattern is to install the Isomer system skills, then ask your CLI agent to operate Isomer through those skills and isomer-cli.

Prerequisites

Isomer Labs depends on Pixi for project and Topic Workspace environments. The recommended Pixi install method on Linux and macOS is:

curl -fsSL https://pixi.sh/install.sh | sh

Paper-writing workflows need a TeX renderer. Tectonic is recommended; a standard LaTeX distribution also works when a project already depends on it. Install Tectonic as a Pixi global tool with:

pixi global install tectonic

Houmao is optional. Install it when you want Houmao-backed agent-team construction, Topic Service Master setup, managed agents, gateways, mailboxes, and related adapter features:

uv tool install houmao

Install

Install the CLI as a uv tool:

uv tool install isomer-labs
isomer-cli --help

For development from a checkout, use Pixi:

pixi install
pixi run isomer-cli --help

Start a Project

mkdir my-isomer-project
cd my-isomer-project
isomer-cli project init
isomer-cli project topics create my-topic --statement "Investigate the concrete research question." --set-default
isomer-cli --print-json project validate

Start the local GUI over the project directory:

isomer-cli project web serve --root .

Then open the printed local URL and choose the Research Topic to inspect.

Topic Workspace Explained

An Isomer Project can contain many Research Topics. Each topic gets its own Topic Workspace, usually under isomer-content/topic-ws/<topic-id>/, and that workspace is the topic's storage and execution boundary.

The Topic Workspace owns the topic Pixi environment. Dependencies for research code, profiling scripts, renderers, and validation tools should be added to the Topic Workspace pixi.toml, not silently installed into the Project root or the user's global Python. This keeps one topic's runtime requirements from leaking into another topic.

The Topic Workspace also owns the research storage layout:

  • intent/ stores the topic overview, environment gate, actor definitions, derived setup gates, and non-canonical user-editable Kaoju template working copies under intent/derived/writing-template/<name>/.
  • records/ stores durable research records, artifacts, evidence items, decisions, view manifests, and the SQLite index used by the GUI.
  • runtime/ stores Workspace Runtime state, adapter payloads, launch material, and resumable execution state.
  • repos/topic-main/ is the topic-owned development repository for code-bearing work.
  • repos/extern/ stores canonical external repositories such as upstream source checkouts.
  • actors/ and agents/ store actor or agent workspaces, usually as worktrees of repos/topic-main.

The Project Manifest records where each Topic Workspace lives, while topic-workspace.toml inside the workspace binds semantic labels such as topic.repos.main, topic.records, and topic.runtime to concrete paths. Agents and skills should resolve those labels instead of guessing paths.

Install System Skills

Install packaged Isomer skills with isomer-cli system-skills install. The core installation unit is the public isomer-op-entrypoint pack. Operator, service, shared-support, and research-recording capabilities, including system-skill management, remain protected members routed by that parent.

Discover optional agent-skill extensions and inspect their entry skills, public commands, and installation guidance before selecting one:

isomer-cli system-skills extensions list
isomer-cli system-skills extensions show kaoju
isomer-cli system-skills install --target codex
isomer-cli system-skills install --target codex --scope user

Supported targets are claude-code, codex, kimi-code, generic, and all. When --scope is omitted, system-skills install defaults to Project scope at the exact current working directory and does not search ancestor Git or Isomer roots. Explicit --scope project is equivalent, while --scope user is the only route to user-wide installation. system-skills status, upgrade, and uninstall require an explicit --scope user|project.

Target Project Scope User Scope
claude-code <cwd>/.claude/skills ${CLAUDE_CONFIG_DIR:-~/.claude}/skills
codex <cwd>/.agents/skills ${CODEX_HOME:-~/.codex}/skills
kimi-code <cwd>/.kimi-code/skills ${KIMI_CODE_HOME:-~/.kimi-code}/skills
generic <cwd>/.agents/skills ~/.agents/skills

Use project scope for skills confined to the current working directory. Choose user scope intentionally when the skills should be available to the selected host across Projects.

Install the DeepSci extension when a project needs hypothesis-driven production research. The selector installs core plus the complete isomer-ext-deepsci-entrypoint pack:

isomer-cli system-skills install --target codex --extension deepsci

Install Kaoju when a project needs evidence-led literature and codebase surveys, first-hand method trials, or controlled comparisons. The selector installs core plus the complete isomer-ext-kaoju-entrypoint pack:

isomer-cli system-skills install --target codex --extension kaoju

The Kaoju pack contains 16 protected members, including paper-search. Paper-search performs bounded provider retrieval through agent-available external tools and records one provider-neutral observation per logical action. Discovery keeps search strategy, candidate selection, and Reading List ownership. isomer-cli ext research literature provides local-only observation recording, derived queries, and explicit index rebuild or validation; it never proxies a literature provider.

Inspect or remove Isomer-owned projections with:

isomer-cli system-skills status --target codex --scope user
isomer-cli system-skills uninstall --target codex --scope user

Direct CLI detection is read-only and inspects only roots supplied by the caller. With no root, it reports declarations and catalog state without scanning provider directories:

isomer-cli --print-json project system-extensions detect
isomer-cli --print-json project system-extensions detect --skill-root /explicit/agent/skill-root
isomer-cli project system-extensions remember kaoju

For newcomer orientation, invoke $isomer-op-welcome; for normal agent-driven operation, use $isomer-op-entrypoint use system-skills to <task>. The protected manager reads Project declarations, current v5 receipts, explicit-root integrity, and limited live inventory in that order. It registers only verified current-v5 complete packs and never removes a declaration because another agent lacks the same installation. Refresh the host or start a new session after installation or upgrade.

Getting Started With CLI Agent

After the CLI and skills are installed, drive Isomer from your coding agent with short skill calls and follow-up prompts.

Create the Research Topic

The user starts with a concrete research goal and wants the agent to turn it into an Isomer Research Topic with durable intent files.

User Action:

$isomer-op-entrypoint use topic-create to create a topic that investigates the concrete research question and defines the expected evidence

AI:

Created the Research Topic, wrote the topic overview and environment gate, and stopped for review.

Prepare the Topic Workspace

The user wants the agent to set up only what the topic needs, verify the environment, and report blockers before research begins.

User Action:

$isomer-op-entrypoint use topic-create to run through setup-topic-env

AI:

Prepared the Topic Workspace, verified dependencies, and reported readiness or blockers.

Run a Bounded Research Pass

The user wants one controlled research pass, not unattended automation, and expects the agent to return artifacts plus a next route.

User Action:

$isomer-ext-deepsci-entrypoint use empirical-pass to run one bounded research pass

AI:

Ran a bounded research pass, recorded artifacts, and reported the next route.

Tighten the Evidence

The user sees that a claim needs stronger evidence and asks the agent to separate what is already supported from what must be measured next.

User Prompt:

Tighten the evidence: compare the model against real measurements, explain mismatches, and revise the next experiment.

AI:

Separated supported claims from missing evidence and proposed a follow-up pass.

Write the Paper

The user wants the accepted records turned into a readable paper artifact, with review blockers surfaced rather than hidden.

User Action:

$isomer-ext-deepsci-entrypoint use paper-pass to draft the paper from accepted records

AI:

Drafted or revised the paper from accepted records, built the artifact, and reported review blockers.

For concrete prompt and response patterns, start with Research Workflow Tutorials.

Documentation

Related Projects

  • academic-research-skills — a collection of prompts and skills for academic research workflows.
  • paper-qa — a framework for scientific literature question answering and review synthesis.
  • deer-flow — a deep research flow framework for autonomous scientific discovery.
  • agent-paper-grounded-reading — an agentic paper reading workflow grounded in source evidence.
  • adhd-paper-reader — a focused, distraction-free paper reader and summarizer.
  • paper-with-code-skills — skills and workflows for reading and reproducing papers with code.

Acknowledgments

The DeepSci skill family originated from the DeepScientist project. Isomer Labs completely refactored those skills to fit its system-skill architecture, Project and Topic Workspace model, artifact contracts, and operating workflows. We thank the DeepScientist authors for their work and inspiration.

Status

The project is early but usable for local research-project setup, topic workspace management, record indexing, and GUI inspection. See Assumptions and Roadmap for current boundaries.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

isomer_labs-0.7.2.tar.gz (6.5 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

isomer_labs-0.7.2-py3-none-any.whl (8.3 MB view details)

Uploaded Python 3

File details

Details for the file isomer_labs-0.7.2.tar.gz.

File metadata

  • Download URL: isomer_labs-0.7.2.tar.gz
  • Upload date:
  • Size: 6.5 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for isomer_labs-0.7.2.tar.gz
Algorithm Hash digest
SHA256 47d4195c599b77fae09c7de670979379403f0bb4e1c045fc5a972b625d213622
MD5 5e983d70ba050e6094d01e755af570fe
BLAKE2b-256 618e532ca8b9c556cdc63c8b93ca51a7a08fb9ab7d1ea3aa0471d321ea8db00c

See more details on using hashes here.

Provenance

The following attestation bundles were made for isomer_labs-0.7.2.tar.gz:

Publisher: workflow.yml on CodeGandee/isomer-labs

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file isomer_labs-0.7.2-py3-none-any.whl.

File metadata

  • Download URL: isomer_labs-0.7.2-py3-none-any.whl
  • Upload date:
  • Size: 8.3 MB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for isomer_labs-0.7.2-py3-none-any.whl
Algorithm Hash digest
SHA256 0efe69e33c0925af98240da0d0eaeef72bf903e66f7a305533b062077934dee0
MD5 82fd9e321c127d098b0c2b76b4383eb2
BLAKE2b-256 fdcbb200f826bac3787d25a4e186cf119d6c3b0e6bd01c84fe788478ea0eecd6

See more details on using hashes here.

Provenance

The following attestation bundles were made for isomer_labs-0.7.2-py3-none-any.whl:

Publisher: workflow.yml on CodeGandee/isomer-labs

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Release history Release notifications | RSS feed

0.8.0

2 files

This release

0.7.2 This release

2 files

0.7.1

2 files

0.7.0

2 files

0.6.2

2 files

0.6.1

2 files

0.5.0

2 files

0.4.0

2 files

0.3.0

2 files

0.2.3

2 files

0.2.2

2 files

0.2.1

2 files

0.2.0

2 files

0.1.2

2 files

0.1.1

2 files

0.1.0

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page