RocketList Startup Jobs

The open data layer for startup hiring. Updated every day by RocketList.

Live roles from funded startups around the world, normalized into one documented dataset for job seekers, researchers, developers, and AI agents.

98,336 active jobs · 7,195 companies · generated 2026-09-20 23:31:05 UTC

Top locations in this snapshot: AB: 1 · AL: 1 · AU: 18 · Albania: 2 · Algeria: 8.

Download

Dataset	Parquet	JSONL	CSV
Jobs	jobs.parquet	jobs.jsonl.gz	jobs.csv.gz
Companies	companies.parquet	companies.jsonl.gz	companies.csv.gz

The full snapshot lives in the stable latest release, not Git history. SHA-256 checksums, generation metadata, and the complete audit result ship beside every snapshot.

wget https://github.com/rocketlist-ai/startup-jobs/releases/latest/download/jobs.parquet

import pandas as pd

jobs = pd.read_parquet(
    "https://github.com/rocketlist-ai/startup-jobs/releases/latest/download/jobs.parquet"
)

berlin_ai = jobs[
    jobs["city"].fillna("").str.contains("Berlin", case=False)
    & jobs["category"].fillna("").str.contains("AI|Data|Engineering", case=False)
]
print(berlin_ai[["company_name", "title", "url"]].head(20))

What is included

The jobs dataset contains factual discovery metadata: company, title, normalized role and seniority, location, compensation when explicitly available, skills, canonical application URL, source platform, and first/last-seen timestamps. The companies dataset adds stage, funding, investors, industry, headquarters, and careers URLs where available.

Schemas are versioned in schema/jobs.schema.json and schema/companies.schema.json. A browsable 100-record sample is committed under sample/, while daily aggregate changes live under changes/. See the public methodology and quality checks for provenance, denominators, audit guarantees, and limitations.

What is deliberately excluded

• Full job descriptions or copied HTML
• Raw ATS responses and crawler payloads
• Embeddings, prompts, traces, or enrichment internals
• Candidate, account, saved-job, application, or matching data
• RocketList's ranking and recommendation logic

The exporter uses an explicit allowlist and the audit fails if a forbidden field appears.

Use it with agents

Install the user-facing RocketList skill in Claude Code, Codex, Cursor, or another compatible agent:

npx skills add rocketlist-ai/startup-jobs --skill rocketlist-job-search

It searches and filters the current snapshot, handles CV-to-role matching, and links users directly to applications. The bundle is also readable at skills/rocketlist-job-search, so any agent can follow it without an installer.

The accompanying distribution loop explains how the dataset, skill, search pages, and recurring data stories compound into discoverability and traffic.

For bulk analysis, give an agent the Parquet URL and the relevant schema. For lower-latency conversational search, connect the RocketList MCP when available:

https://rocketlist.ai/mcp

Its official MCP Registry manifest is versioned in server.json.

Example prompt:

Use the RocketList dataset to find active Series A–C companies in Berlin hiring product managers. Return the canonical application links and explain the filters you applied.

Update model

The workflow runs daily at 04:17 UTC:

1. Fetch all current public catalog rows through paginated API reads.
2. Normalize join keys, field types, lists, and stable public IDs.
3. Retain active, non-duplicate jobs and active or referenced companies.
4. Cross-foot totals, reconcile API counts, verify unique IDs and URLs, test company references, and reject forbidden fields.
5. Replace the full assets on the stable latest release.
6. Commit only small samples, statistics, and daily change summaries.

No credentials are required to reproduce the export:

python -m pip install -r requirements.txt
python scripts/export_dataset.py
python scripts/audit_dataset.py

Accuracy and limitations

RocketList aggregates company career pages and ATS sources. A listed role can close between daily refreshes; the canonical application page is authoritative. Coverage varies by employer, country, and field, and missing values are never imputed for public statistics. See stats/latest.json for field-level denominators.

License and attribution

The dataset is available under ODC-BY 1.0; code is MIT licensed. Attribute RocketList with links to https://rocketlist.ai and this repository. Employer names and trademarks belong to their respective owners, and source postings remain subject to their publishers' terms.

Corrections

Open an issue for a missing company, stale role, broken URL, or schema problem. See CONTRIBUTING.md for the data-safety rules.