flow8 runs AI agents, private knowledge bases, and document intelligence on your own infrastructure. The model reasons, your flows execute — and every single step lands in the audit trail.
A general-purpose assistant can draft a reply — but it can't check the customer record, create the credit note, or update your accounting. The work stays manual.
Cloud agent builders route your prompts, documents, and credentials through someone else's infrastructure. For regulated teams, that's a non-starter.
When an agent acts without a record, compliance can't sign off. Most platforms keep a chat transcript at best — not an audit trail.
flow8 agents are different by architecture.
Agents run inside the execution layer — on your infrastructure, with your flows as their tools, and the same immutable audit trail as every other process.
An agent in flow8 doesn't improvise against your systems. It reasons with the model you choose and acts through flows you've built — look up a customer, create an invoice, escalate a case. What the agent may touch is exactly what you grant. Nothing more.
Via MCP, the loop works in both directions: agents inside flows call your processes as tools, and external assistants like Claude, ChatGPT, or ChatGPT Codex can invoke them too. Bring your own model — self-hosted or cloud, switched in settings.
Build knowledge bases from contracts, policies, manuals, and records — chunked, embedded, and searched entirely inside your infrastructure.
Text chunking, embedding generation, and vector upserts are native flow8 modules. No Python glue, no separate pipeline service, no exports to a third-party RAG product.
Qdrant and Weaviate run next to your data. Your documents and their embeddings never leave your network — the index is as sovereign as the source files.
Vector similarity is combined with BM25 keyword scoring and merged via reciprocal rank fusion — so retrieval stays precise even on contract numbers, part codes, and legal jargon.
Source tracking is built into the chunker — every answer points back to the documents it came from, and the retrieval itself is logged like any other step.
The unstructured mess that arrives every day — PDFs, scans, recordings — leaves as clean, validated data.
Invoices, contracts, and forms become clean JSON validated against your schema. No regex archaeology, no manual keying.
Scans, photos, and screenshots analyzed by vision models — classify damage reports, read submitted documents, check ID photos.
Calls and meetings transcribed inside the flow — then summarized, routed, and filed automatically.
Drafts, summaries, and images generated mid-process — from response letters to report graphics.
Every AI step is recorded like any other step in the execution log — which model ran, what went in, what came out, how long it took. When compliance asks how a decision was made, the answer is one query away.
An agent reads the invoice, validates it against the order, posts to accounting — and escalates to a human only when something doesn't add up.
"Which agreements auto-renew next quarter?" — answered from your own contract archive, grounded in the actual clauses, inside your network.
Incoming requests classified, answered from the knowledge base, and escalated with full context when a human needs to step in.
Recordings transcribed, decisions extracted, tasks created in your systems — minutes after the call ends.
IDs and submitted forms verified by vision models running inside your network — never via a third-party verification API.
HR and IT policies answerable in Slack or Teams — grounded in the current document version, not office folklore.
Book 30 minutes. Bring a process and a folder of documents — we'll show an agent executing it while your data stays exactly where it is.
Self-hosted models · Your flows as tools · Every step audited