cpe-data-cli 0.4.0

Read-only CLI for a research data API (wiki, graph, memos, raw, coverage, eval). Self-documenting; an agent can drive it without a skill.

data-cli

CLI for a research data API. Gives an agent (or you) access to every research surface — memos, company wiki, knowledge graph, raw primary-source files, coverage, and eval — to analyze a stock, and can trigger research (gather a ticker, refetch a source, comment on a memo) through one self-documenting tool.

The key grants read access to every surface plus a small set of action POSTs (gather/new symbol, re-run a phase, refetch/repair/fetch-missing a source, refresh people radar, comment on a memo). It cannot upload externally-built content, edit events, run model bakeoffs/simulations, or run grounded chat — those return 403. It is an actions API, not a content writer.

This tool talks to a private data API. You need two values, provided to you separately: the API base URL and a key. Without both, the tool does nothing.

Install

pip install cpe-data-cli
# or, isolated:
pipx install cpe-data-cli

Upgrade

pip install --upgrade cpe-data-cli
# or, with pipx:
pipx upgrade cpe-data-cli

Configure

export CPE_API_BASE=<base url you were given>
export CPE_API_KEY=<key you were given>
data-cli auth check              # verify base + key + connectivity

Both are required. The tool makes no network call until both are set.

Version & completion

data-cli --version               # or -V
data-cli --install-completion    # install shell completion (bash/zsh/fish)

Analyze a stock

Read the synthesized dossier, then pull the specific primary-source figures you want to cite. Every command prints the next commands to run.

data-cli memo get NVDA                  # START HERE: the full verified dossier (markdown)
data-cli wiki get NVDA                  # synthesized company page (business, segments, model)
data-cli ticker NVDA                    # one-line orientation (latest run, coverage%, raw count)
data-cli raw NVDA source_map.json       # provenance map: read FIRST to find Tier-1 files
data-cli raw NVDA latest_10k.json       # a specific primary source (10-K, financials, ...)
data-cli graph ticker NVDA              # supply/customer/competitor/segment web

More surfaces:

data-cli memo list NVDA                 # other runs/dossiers for the ticker
data-cli wiki list                      # all covered companies
data-cli raw NVDA                       # list every primary-source file
data-cli coverage NVDA                  # source-completeness (check before quoting a number)
data-cli eval NVDA                      # quality verdict + contradiction tally
data-cli graph search "TSMC"            # find nodes by NAME (companies/people, not concepts)
data-cli graph all                      # the full knowledge graph
data-cli guide                          # this workflow, any time

Read the trust layer as a citation check, not the story: build the analysis from the memo + Tier-1 sources, glance at coverage/eval before quoting a specific number, and don't lead with contradicted claims.

Trigger research

These spend real compute; use them deliberately.

data-cli gather NVDA                     # gather/refresh a ticker (works for a new symbol too)
data-cli gather NVDA --model opus        # pick a model
data-cli gather NVDA --phase analyze     # re-run a single pipeline phase
data-cli refetch NVDA financials         # refetch one source category
data-cli fetch-missing NVDA              # fetch all missing sources
data-cli repair <source_name>            # repair a specific source
data-cli people-radar refresh            # refresh the people radar
data-cli comment <RUN_ID> --section <s> --issue-type <t> --note "..."   # comment on a memo

Output

• Markdown-native surfaces (wiki, memo, raw .md) print markdown by default. coverage and eval print a human-readable summary by default (the trust layer, scannable at a glance); add --json for the full payload. Other structured surfaces (ticker, graph, lists) print JSON.
• --json on any data command for structured output (with _next breadcrumbs in the body).
• -o FILE writes the content to a file (confirmation goes to stderr). Plain > file.md works too — diagnostics and breadcrumbs go to stderr, so a redirected or -o file holds only clean content. (auth check and guide are diagnostics and ignore -o.)

For AI agents

This tool ships with an agent skill that teaches the right way to drive it (analyze a stock from the memo + Tier-1 sources, read the trust layer as a citation check, and the common pitfalls):

data-cli skill          # print the skill
data-cli skill --path   # print its path — point your agent's skills dir at it