ragweld
A tribrid RAG workbench: benchmark, debug, and ship.
Hybrid retrieval (vector + sparse + graph) with the engineering surface area you actually need in production.
Live search + chat are backed by the open-source Faxbot + Vivified corpora. Add ?mock=1 for offline demo mode.
What ragweld actually is
The retrieval stack matters — but ragweld’s differentiator is the engineering surface area around it: benchmarking, evaluation, diagnosis, and operations.
It’s built for senior engineers who need answers like: what changed, why did it change, and what should we try next — without guessing.
Hybrid + graph, fused and inspectable
Tribrid retrieval: vector + sparse + graph legs. Fuse signals, inspect sources, and tune each leg instead of treating retrieval like a black box.
Measure regressions, not vibes
Run benchmarks and eval suites, then compare runs to see what changed — with drilldowns and evidence instead of guesswork.
Diff the “why”
Inspect per-query outcomes, retrieval legs, and sources. Diff runs across 200+ parameters (down to embeddings + tokenization), and get a one-click comparison summary of what likely mattered.
Learning loop built in
Cross-encoder reranking with a feedback workflow: capture tuples, mine triplets, and compare/promote runs. Automation is first-class (cron/CI) because production doesn’t wait for a button click.
A built-in RAG wiki
Verbose explanations for knobs and tradeoffs — often with direct links to papers, repos, and third-party tools.
See metrics while you run
Grafana embedded in the UI (split-screen) so you can run a query, eval, or benchmark and watch live metrics in the same workspace — plus alerting hooks for production ops.
Enterprise-grade, configurable
Alerting isn’t “on/off” — it’s your rules, your on-call. ragweld provides the plumbing: Slack/Discord webhooks and Alertmanager visibility so production incidents don’t hide in logs.
A full chat interface, too
Multi-model chat (including vision inputs), persistent memory (sparse + dense), multi-corpus querying, and per-corpus embeddings/tokenization when your data demands it.
Use it anywhere
Built-in MCP server support so you can use TriBridRAG tools from other clients (IDEs, agents, automation) without rewriting integrations for each surface.
Under the hood
Tribrid retrieval is the spine: vector + sparse + graph signals fused and reranked — then surfaced through a workbench that lets you measure and iterate.
Choose your path
ragweld is open source and self-hostable — and we also support enterprise deployments.
Self-hosted / on-prem
MIT licensed. Run it on your infrastructure. Keep your corpora, embeddings, and model traffic where you want them.
- • Tribrid retrieval: vector + sparse + graph
- • Benchmark + eval suites with run comparisons
- • Deep config + parameter glossary + tooltips
- • Multi-corpus + per-corpus overrides
Managed deployments + custom integrations
For teams pushing RAG into production: deployment patterns, hardened ops, and integration work tailored to your stack.
- • Cloud or on-prem deployments (your network, your controls)
- • Observability + alerting hooks that plug into your on-call
- • Benchmarks/evals as a release gate (regressions don’t ship)
- • MCP + automation workflows for agents and internal tools
Ready to see ragweld in action?
Explore the live demo, skim the docs, and then talk to us if you’re deploying this in production.