AI workforce platform

Hand off the paperwork.
Keep your team.

Specialist AI agents handle the research, invoice review, compliance scans, and code maintenance that drains your day. They run on your own infrastructure, audit every action, and cite every claim.

Start free trial Self-host with Helm →

Source-available · commercial license required to run in production · Get a license

What you can hand off

Six things your team would rather not do today.

scholar

Weekly competitor brief

Crawls product blogs, GitHub releases, and Hacker News every Monday at 8. Emails a summary with citations to the team.

critic

Sort and label invoices

Watches the shared cloud folder for new PDFs. Extracts vendor, amount, and date, files them into the right project.

developer

Banner for next week's event

Edits the homepage repo, adds the banner with a link, runs the build, opens a diff in the cockpit for review.

critic

Compliance scan on contracts

Weekly check on the contracts folder. Flags missing GDPR clauses, expired NDAs, and counter-party changes.

scholar

Daily SQL summary

Runs queries against your warehouse each morning, writes a one-page revenue brief, emails it before standup.

secretary

Triage support emails

Parses the support inbox, sorts by topic, files into the right project thread, drafts a reply for review.

How it works

Submit. Work. Review.

1

Submit a job

Cockpit, REST API, MCP, or cron schedule. Pick an expert role and attach what it needs — a cloud folder, a Postgres datasource, a Git repo, a Slack channel.

2

Agent works in isolation

The agent picks up the job in a dedicated SSH-accessible workspace. It plans, calls tools, writes intermediate state, and pauses when it needs human input.

3

Review what it did

Outputs land in your cloud folder. Diffs open in the cockpit for approval. Every tool call and LLM request is in the audit log.

In the cockpit

Two views you'll spend the most time in.

Diff review

Every change the agent makes opens as a diff. Approve, reject, or send feedback before anything lands in your cloud folder or your repo.

Job dashboard

Supervise the fleet from one screen. Status, expert, and progress at a glance. Click any row for live shell, audit, and the agent's plan.

Why this, not a chatbot

What ChatGPT, copilots, and RPA can't give you.

Beyond RPA

RPA automates clicks. This hands off the work behind them. Agents read context, decide, take action, explain what they did.

Workspace per job

Every run gets a dedicated SSH-accessible container or VM. The agent never touches your servers directly, and you can audit what it touched.

Append-only audit trail

Every tool call, every LLM request, every output — recorded in MongoDB, queryable by job, agent, or time range.

Citations on every claim

Factual claims link back to the source the agent actually read. No hallucinated answers you can't trace.

Specialists you can fork

Scholar, developer, critic, curator, designer — each defined in a YAML config. Diff and PR new roles like code.

Scheduled and event-driven

Cron triggers with timezone and DST support. Webhook triggers next. Long-running jobs that don't need a human in the loop.

Scale

From a minipc to a multi-cluster fleet.

MINIPC orchestrator × 1 agent × 1 postgres · neo4j mongo · redis One team, a minipc. SINGLE VM orchestrator × 1 agent × 3 postgres · neo4j mongo · redis A department, a single box. K8S FLEET orch × 3 HA agent × N (autoscale) postgres · neo4j · mongo replicated, multi-zone Thousands of employees, a fleet.

Same Helm chart, same code. Add agent workers by changing one number in values.yaml. Run on a minipc for one team, or scale to a multi-cluster fleet for thousands of employees.

Two ways to run it

Same codebase, different operating model.

Hosted

Managed subscription. We run the orchestrator, the database cluster, the agent fleet, and the upgrades. You sign in and use it. Best when you don't want to operate Kubernetes yourself.

Start free trial See pricing →

Self-hosted

Your cluster, your data. Ships as a Helm chart with a terms-of-service gate where you attest you hold a valid commercial license — same model as Neo4j Enterprise. Reading the source is free; running it commercially is not.

Read the Helm guide Get a license →

SRW is source-available, not open source. The repo is public so you can audit it, fork for testing, and verify what runs in your network. To run it in production — hosted by us or by you — you need a paid subscription or a commercial license. We'd rather tell you here than at the Helm gate.

Built on

FastAPI · LangGraph · PostgreSQL/pgvector · MongoDB · Neo4j · Keycloak · Redis · Angular · Helm · Kubernetes