I.
Multi-agent, by design
Researcher, reviewer, verifier, writer, contradiction - 20+ specialised subagents coordinated by an orchestrator.
00 | INDEX - BOHR RESEARCH LABORATORY | LOCAL SESSION - PUBLISHED MMXXVI
Bohr AI
v0.2.18
VOL. II | NO. 17 - CLI RESEARCH SHELL
A multi-agent research shell that assembles, critiques and verifies a paper the way a lab does - from literature sweep to peer review, in your terminal.
-- FIG. 01 | LIVE SESSION /DEEPRESEARCH - ~/research/transformers
A real run, replayed.
Click a sample or type your own workflow prompt
BOHR ~/research/transformers
Live
PID 18332 - CLAUDE-OPUS-4 - 25 TOOLS
# bohr v0.2.18 - CLAUDE-OPUS-4 - 25 TOOLS - host: local machine
# workspace: ~/research/transformers - session: preview-transformers
bohr> /deepresearch optimization techniques for transformer architectures
> PLAN research-planner scoped 7 stages | 4 researchers | stop@conf>0.85
> FAN-OUT AlphaXiv, web, and code search dispatched in parallel
> CHECK contradiction agent found 2 efficiency tradeoff disputes
> VERIFY verifier anchored 87/87 citations before final promotion
FILE outputs/transformer-optimization.md | 4,281 words | 87 citations
FILE outputs/transformer-optimization.provenance.md
- 01 | PILLARS
Research, not autocomplete.
Bohr AI is built for people who write papers, run replications and defend claims. The shell orchestrates specialised agents, verifies against sources, and leaves a trail - not a chat log.
II.
Cited & verified
Every claim is anchored to a source; a verifier agent cross-checks citations against their text before delivery.
III.
CLI-first
Nothing depends on a proprietary dashboard. Artifacts land as markdown in your repo, versionable with git.
IV.
Adversarial loop
Findings are stress-tested by a contradiction agent and a reasoning validator until convergence criteria are met.
- 02 | WORKFLOWS
Eleven slash commands, one shell.
Each command is a full pipeline - planner, researchers, scorer, validator, visual asset builders, verifier, reviewer - with agents and tools dispatched in the right order.
All workflows ->- /deepresearch 01 Durable research brief with inline citations, provenance, and verification.
- /lit-review 02 End-to-end literature survey with themes, contradictions, gaps and validation.
- /paper 03 Publication-grade manuscript with method, figures, citation & compliance passes.
- /visuals 04 Data-backed charts, source-linked tables, and conceptual image assets.
- /review 05 Simulated peer review with severity-graded feedback and revision guidance.
- /audit 06 Compare a paper's claims against its public codebase for reproducibility.
- /replicate 07 Replication plan or workflow for a paper, claim, or benchmark.
- /compare 08 Agreement / disagreement matrix across sources on a topic.
- /autoresearch 09 Autonomous experiment loop that iteratively optimizes toward a stated goal.
- /draft 10 Turn findings into a structured, multi-step manuscript draft.
- /watch 11 Recurring monitoring on a topic or literature area.
- 03 | RESEARCH-GRADE | NOT A FREE ESSAY BOT
Grounded workflows, not one-shot prose
Generic AI writers optimize for fluent paragraphs in one chat. Bohr runs multi-agent pipelines with citations, verifier passes, and artifacts on disk - so tokens go to evidence and synthesis, not blind regurgitation.
Works with your provider
OpenAI Anthropic Google xAI DeepSeek + OpenRouter & more
Typical free writer
- Single model, single thread
- Sounds academic; weak traceability
- Re-prompt until it "feels" right
Bohr
- Slash workflows & specialised agents
- Provenance, verification, compliance passes
- File handoffs & token-aware subagents
Fewer filler tokens Grounded citations Cost docs
- 04 | INSTALLATION
One line.
Your terminal.
Bohr runs locally, with optional Docker. Add your model keys to .env and you're in the REPL.
curl -fsSL https://raw.githubusercontent.com/Dhruvil7694/BohrAI/main/website/public/install | bash Prefer a walkthrough? Read the installation guide -> Clone on GitHub ->