Primr

Company name + website → evidence-grounded, long-form strategic reports. Built for agent hosts and the CLI: structured research, strategic insight, confidence labels, and durable Word and Markdown artifacts, not a free-form chat essay.

Primr gathers public site pages, DNS/recon, hiring signals, and other open sources, develops a strategic view of the company, then produces a consultant-style Strategic Overview and selected YAML-defined strategy documents with citations and Confirmed / Reported / Estimated / Hypothesis labels. Point an agent at the repo, or run the CLI. Both paths use the same evidence, uncertainty, and long-form artifact contract, but their execution, model ownership, and assurance are different.

<p align="center"> <img src="docs/images/primr-demo.png" alt="Illustrative primr CLI session for ExampleCo: estimate, then report artifacts" width="920" loading="lazy" /> </p>

<p align="center">_{Placeholder demo data (ExampleCo), not a live company capture.}</p>

primr "ExampleCo" https://example.co

In an agent chat, that request defaults to Primr Zero: keyless evidence collection in Primr, research and writing in the host (when the host plan does not bill API usage). When a human runs the command directly in a terminal, it retains the provider-backed CLI path after a dry-run estimate and approval.

What it is for

Use Primr when you need a serious first draft for discovery, account planning, diligence, competitive analysis, or strategy work, not a two-paragraph pre-call blurb (use normal web search for that).

• Structured brief with uncertainty labels, not scattered notes
• Research grounded in public evidence, not only search summaries
• Cost-aware local execution: dry-run before billable work
• Reusable Markdown/DOCX artifacts for humans, agents, and downstream tools

Primr is not a generic crawler, a SaaS collaboration app, a model-serving platform, or a tool for bypassing authentication, paywalls, or site restrictions.

Start here

Both paths can deliver a Strategic Overview and AI Strategy as Markdown and DOCX (on the Zero path, primr render <file>.md converts host Markdown to DOCX at $0). The provider-backed path additionally owns its measured synthesis, cross-validation, usage accounting, and recovery stages. A configured API key is capability, not consent to spend.

Primr prefers a validated zero-incremental-cost route when the current agent host or local environment can support the full artifact contract honestly. When provider APIs are required, the Standard recipe targets the best complete report and default strategy artifacts it can produce for approximately $1; the exact dry-run estimate remains authoritative and execution still requires explicit approval.

Agent path

Point a capable agent at this repository:

primr "ExampleCo" https://example.co

The agent uses Primr Zero unless you explicitly request paid, metered, provider-backed, or premium execution. Its primr prep collection is already noninteractive and needs no approval flag. Routing, confirmation rules, MCP setup, and host handoff live in Agent Integration and Zero-Cost / Primr Zero.

For a smaller draft using an installed local model, run primr draft --local BUNDLE --dry-run, then remove --dry-run to execute. This also supports an explicitly configured private-network inference box. The draft records evidence coverage, model, quantization, runtime, and zero model API spend. It does not run Primr's external research or validation stages; review each claim against its displayed source excerpt. See local draft setup and limits.

Terminal path

primr "ExampleCo" https://example.co --dry-run --budget 10
primr "ExampleCo" https://example.co                  # foreground, then approve
primr "ExampleCo" https://example.co --skip-confirm   # automation, after approval

--skip-confirm is the approval signal for a noninteractive or background launch of the standard provider-backed commands. Use it only after a person has reviewed and approved the fresh quote. The experimental primr orchestrate command is the exception: its noninteractive approval and spend ceiling is --max-cost <usd> after the exact dry run is approved.

Mode matrix, platforms, strategy types, and cost controls: Run Modes and Costs.

Install

• Python 3.11 through 3.15; Python 3.15 is the default
• No API key or GPU for primr recon / primr prep
• Keys only for provider-backed research (preferred default: OpenRouter gateway when configured, or direct xAI + Gemini including Google Gemini 3.8 Flash; optional direct OpenAI, Anthropic, Bedrock, Azure)
• primr init installs browser deps for scrape tiers

uv tool install --python 3.15 primr
primr --version

With pipx, use pipx install --python python3.15 primr and an installed Python 3.15. Plain pip install primr also works inside a Python 3.11 through 3.15 virtual environment. Download and inspect convenience installers before running them. For those scripts, upgrades, PATH troubleshooting, and source-checkout setup, see Installation and Upgrades.

Provider-backed setup only when you want billable runs:

primr init
primr doctor

Keys and full config: API Key Setup | OpenRouter Gateway | Configuration.

Common commands

Focused help: primr --help. Everything: primr --help-all.

Cost gate

Billable runs need a fresh --dry-run estimate and explicit approval. A --budget 10 cap sets a $10 ceiling for that Primr run, not a $10 target or permission to launch. The preview reports whether the estimate fits, and an oversized plan is not launch-ready. Unattended execution stays fail closed unless it carries the command-specific approval signal. See Run Modes and Costs for the complete policy.

Outputs

Default: customer-facing files under output/, diagnostics under working/.

• <Company>_Strategic_Overview_<date>.md / .docx
• <Company>_AI_Strategy_<date>.md / .docx (unless --no-ai-strategy)
• run_manifest.json (estimate, approval, audit)

Agent inventory (paths/roles only, no report body): primr://output/artifacts/by_job/{job_id}. Full artifact and recovery guides: Artifacts · Recovery.

Docs

Docs site index: docs/README.md.

Disclaimer

Primr retrieves and analyzes public web content. You are responsible for site terms, provider costs, output accuracy, and legal fit. AI-generated analysis can be wrong or outdated. Verify important findings before acting on them.

License

Apache 2.0. See LICENSE.