// platform · decisions for AI engineering

From a change to a measured decision

Outrider drafts promising changes. Validate measures them on the pull request. Together they create the record for the next decision.

Try Validate See pricing

// 01 capabilities

What connects a change to its result

recommendations

Candidates drawn from research and matched to your repository context. Outrider checks fit before drafting implementation.

research_interests

A research brief for what to track. Create one from free-form context, a GitHub repo, or a project's experiments, and the engine matches new work to it daily. In the product this brief is a Research Interest.

experiment_records

Validation runs keep the reviewed plan, evidence, verdict and cost linked to the pull request. Studio lets you inspect those records.

validation

On a connected pull request, Validate proposes the test, runs the approved comparison on Remyx compute and reports back to GitHub.

decision_records

Validate returns a verdict against the approved test. Maintainers review the evidence and decide whether the change should merge.

portfolio_view

Product direction: connect results across repositories and AI systems to inform the organization's next investment.

// 02 decision_records

Every validation leaves evidence the team can revisit

A PR, approved test, evidence, verdict, and cost stay together. The team keeps the merge decision.

evaluation plan, change, and results, linked
the decision, with its rationale
run history that stays available in Studio

PR #412 · proposed reranker CLOSED · 2 DAYS

evaluation planAdding a reranker improves answer resolution within the latency limit.

changePR #412 rerank top-20 → top-5 before context build

resultsanswer resolution +2.6 pts (target +2.0) · p95 +64 ms, limit +80 · cost per 1,000 queries +17%

verdictSHIP quality target and latency limit pass; cost rise accepted, watch on long contexts

# illustrative example

// 03 patterns

See the potential of cross-run evidence

The next layer groups outcomes by approach. The cards below illustrate the intended view, not live customer data.

retrieval

5 of 5 positive, avg +3.2%. HIGH SIGNAL

prompting

2 of 2 positive, avg +1.2%. MODERATE

routing

0 of 2 significant. LOW SIGNAL

# illustrative example

changedchangeverdict

Mretrieval/reranker.pyPR #412SHIP

Mtools/schema.jsonPR #405SHIP

Mprompts/system.mdPR #398ITERATE

Mrouter/fallback.pyPR #391REJECT

Mcontext/builder.pyPR #387SHIP

# illustrative example

// 04 the_loop

A record today. Better direction over time.

Evidence history

What was considered, tested, measured and decided.

Hypothesis prioritization

Use previous results and team decisions to identify promising directions for further evaluation.

CURRENT · ONE PULL REQUEST

Proposed changeReviewed testVerdict + evidenceActual cost

RETAINED RECORDWhat was tried · What happened · What the team decided

LONG-TERM DIRECTIONUse outcomes across systems to guide the next recommendation and investment

// 05 the_remyx_framework

From one change to an AI development strategy

Measure a change today. Build a record for where to invest next.

Evaluate changes

What changed? Did it meet the criteria in the plan? What tradeoffs came with it?

→

Learn from results

Keep evaluations, cost and team decisions together. Bringing in production outcomes is part of the roadmap.

→

Guide development

Which approaches are worth testing next? What should the team try next? Where does the evidence change the decision?

Next change

The cross-system view is a longer-term platform goal, built from repeatable evidence on individual changes.

// 06 integrations

Drops into the stack you already run

Start in GitHub. Outrider runs in Actions. Validate runs through the Remyx GitHub App. Settings and billing live in Studio.

# bring your own provider key: Anthropic, z.ai GLM, Moonshot Kimi, OpenRouter, Bedrock, Vertex, and more.

Measure the change in front of you

Connect the GitHub App and comment @remyx-ai validate on a pull request.

Try Validate See pricing