Riviera lets you fix how agents use your API without redeploying the backend
Rodrigocruuz · reddit · 2026-10-09
Developer Rodrigocruuz launched riviera.so, addressing a common agent failure mode: an agent picks the wrong tool or passes wrong params, the request still succeeds, logs look fine, but the user didn't get what they asked for — usually because tool descriptions are vague.
How it works:
- Import an OpenAPI doc or connect an existing MCP server
- Edit the tool and parameter descriptions agents read (when to use the tool, where IDs come from, dollars vs cents)
- Changes stay in draft until you publish a version; backend execution unchanged
Useful scenarios:
- Disambiguating similar tools ("create invoice" vs "record a payment already made")
- Adding parameter context (ID sources, units, date formats)
- Updating descriptions independently of deploys
- Debugging: see which tool was called, published version, duration, outcome
- Turning mistakes into tests: save a request and expected tool, compare production vs draft
Limitation: model tests aren't self-serve yet; they check tool selection without executing backend calls.
More from coding & agent
- Professor delegates multi-step task to AI agent Sai, which nails it in one shot — xwang_lk · 2026-10-09
- Dev Has Codex Build Its Own Frame Profiler to Auto-Optimize a Custom Game Engine — Dimillian · 2026-10-09
- Open-Source AI Skill Replicates Nikita Bier's $500-a-Minute Growth Playbook for Free — FinanceYF5 · 2026-10-09
- 40k-token tool outputs vs small agent contexts: Reddit weighs truncation, summarization and handles — No-Age-3362 · 2026-10-09
- Comma: open-source 24/7 personal agent that runs background tasks to completion — FellMentKE · 2026-10-09
- 20 agents, 800k test executions: the hard part is proving agents did what they claimed — Grimmoner · 2026-10-09