— PROCESS · OVERVIEW

How it works

Bohr is CLI-first: you drive everything from the REPL with slash commands. Orchestration coordinates agents and external tools; nothing depends on a proprietary web dashboard.

- PROCESS RAIL

Five visible stages from setup to artifacts

This should feel like a workflow, not a checklist. Each step shows the handoff happening in the shell: setup, dispatch, orchestration, and files on disk.

CLI first Agents + tools Artifacts on disk

  1. Machine setup

    Installation guide ->

    Install and configure

    Install the Bohr CLI, copy `.env.example` to `.env`, and add model provider keys. See Installation for OS-specific notes.

    Prepare the shell

    bohr install .env provider keys

  2. Start a session

    Quickstart ->

    Open a session

    Run `bohr` (or your wrapper). You land in the Pi-powered REPL where slash commands map to full workflows.

    Enter the REPL

    terminal bohr session ready

  3. Pick an outcome

    Browse workflows ->

    Choose a workflow

    Pick by outcome: `/deepresearch` for cited briefs, `/lit` or `/lit-review` for surveys, `/paper` for manuscripts, `/audit` for claim-vs-code checks - see Workflows.

    Dispatch a command

    /deepresearch /lit-review /paper

  4. Orchestration

    Meet the agents ->

    Let orchestration run

    The orchestrator sequences agents (researcher, writer, verifier, and more), runs tools, and writes intermediate files under your workspace conventions.

    Fan out work

    planner agents tools

  5. Inspect outputs

    Preview docs ->

    Review artifacts

    Outputs land as markdown and sidecars (e.g. provenance). Use `/preview` when configured, or open files in your editor.

    Audit the result

    brief.md provenance notes

Diagrams

Same staged pipeline visuals as workflow pages (grid, outlines, hero blocks). Left: how a session flows from install to artifacts. Right: the /deepresearch diagram — identical data to the Deep Research workflow page; your run may merge or repeat stages depending on planner and iteration policy.

Session flow

Left-to-right: install and keys, REPL + slash commands, orchestration calling agents and tools, then files on disk.

Example · /deepresearch pipeline

Stages and labels match the workflow hub; arrows show how work moves across the pipeline (same renderer as Flow diagram on workflow pages).

- DECISION GUIDE

Which workflow should I use?

Pick by the artifact you want, not by vendor. Bohr runs in your environment; the command choice is really an output choice.

brief survey manuscript stress test
01 Cited brief

I need a cited brief on a messy topic

Start here when the question is broad, the sources are mixed, and you need a defensible answer with references.

/deepresearch
Open guide
02 Survey map

I need a literature survey with gaps and disagreements

Use this when the goal is field coverage: themes, contradictions, consensus, and missing work.

/lit-review
Open guide
03 Manuscript draft

I need a paper-style draft with compliance checks

Best when you already know the paper you want and need drafting, figures, compliance, and verification passes.

/paper
Open guide
04 Adversarial check

I need to challenge conclusions against evidence

Run this when the claim exists already and you want the strongest counter-evidence, gaps, or reasoning failures.

/contradict/reasoning-validate
Open guide

Next