No SDK, no dashboard, no protocol to learn. grepapi is a CLI: JSON on stdout,
typed errors on stderr, semantic exit codes. Your agent runs grepapi guide
once and drives the whole thing — find, rank, enrich, brief — then writes the outreach itself.
Agent-native by design
The CLI is the interface
JSON on stdout, typed errors on stderr, exit codes (0 · 80 input · 90 precondition · 100 retryable). Branch on failure, pipe the output.
grepapi guide
A version-exact skill: the model, loop, sources, actions, enrichment, recipes, examples. Your agent reads it once and self-onboards.
grepapi doctor --fix
Self-healing: it sets up the fetch environment so the agent doesn't have to debug the browser.
BYO-LLM
The operator is the LLM. grepapi returns a brief; your agent writes the copy. No completions run server-side, no token markup.
The loop your agent runs
# learn the tool, set up a hunt, work it — all as JSON grepapi guide # {model, loop, concepts, actions, examples} grepapi recipe find-a-partner "…" # or compose a custom ICP grepapi run # fetch (its browser) → scored, enriched leads while :; do n=$(grepapi next) # {action, lead_id, brief:{lead, playbook, instruction}} [ "$(jq -r .lead_id <<<"$n")" = null ] && break # … the agent writes the message from $n.brief … grepapi next --done # sent — advance to the next best done
Every lead comes back with a who, a channel (call · comment · reply · email), and a brief — freshness- and outcome-aware, so next surfaces what converts.
What your agent gets
One install line, a self-teaching guide, and a free tier. Your agent does the lead-gen you'd rather not.
# one line curl -sSL https://grepapi.intrane.fr/install.sh | sh