Read Bright with an agent.

Agent contract: 1.2.0

Bright connects documented claims to primary sources, limitations and interactive experiences. Every representation comes from the public content records. No account, key or cookies are needed.

Connect an assistant

Assistants that support remote MCP connectors can add Bright at https://brightaifuture.com/api/mcp. It is read-only and needs no sign-in. See Ask Bright for setup, example requests and the tool list.

Choose an operation

getBrightGuide — explain Open Models or Data Centers with an answer, caveats and evidence.
exploreBrightOpenModels — find model families, plan RAM/VRAM/unified-memory budgets, or compare openness.
discoverBrightRecords — find records and documented developments; use a stable ID to inspect their evidence.
getBrightWeekly — read the latest approved Bright Weekly at /api/weekly, one edition by id, or the list with list=true.
findBrightClaimEvidence — find records whose evidence bears on a claim at /api/claims?claim=…. It returns related evidence and limitations, never a verdict.

Ask for context when the subject is unclear. Bright is not a general-purpose answer service. Specific local outcomes require local records; a hardware estimate is not a tested configuration.

Find a claim and inspect its evidence

Search the readable site, or GET /api/discovery?q=speech.
Retrieve one stable ID: /api/discovery?id=speech. Read its documented result, original sources, limitations, review and linked experiences before answering.
Use the canonical URL when citing Bright. Cite the primary source for the underlying finding; distinguish observed results from possible implications.

Open models and data centers

For Open Models, explain open weights versus open source, inspect model licenses and provenance, explore local or private inference, find families by use case, estimate a hardware weight budget, or compare the nine openness dimensions. For Data Centers, examine electricity bills and demand, water withdrawal versus consumption, noise and humming, construction versus permanent jobs, taxes, grid infrastructure, backup power and community agreements. Keep the evidence and limits attached.

Ask a guide question

Use the bounded guide API for explanatory, planning and locality-aware questions. It accepts one of id, slug, q or query, with optional topic and geography. Responses start with an answer-first synthesis and keep sections, caveats, evidence, canonical links and review dates together.

Open source vs open weight — terminology, what is released and what remains unknown.
Data-center electricity in Virginia — a geography-aware example; local tariffs, utility rules and infrastructure records matter.

Explore and compare open models

/api/models provides bounded planning and comparison views over the checked family atlas. Explore with action=explore and an atlas intent such as code; filter with modality, size, commercial and reproducibility. Compare two to five family IDs while retaining each openness dimension:

Hardware results are plausible planning candidates, not compatibility guarantees. Runtime, backend, context length, quantization, offload, overhead and workload can change what fits. Model licenses and openness dimensions are reported per family; Bright does not collapse them into a score or select a universal winner.

Representations

RSS · JSON Feed · OpenAPI 3.1 · llms.txt · Example Markdown record · Sitemap

Pages advertise Markdown through alternate links and HTTP Link headers. Markdown lives at /read followed by the canonical path and .md; the home representation is /read/index.md.

Read-only API contract

GET /api/discovery accepts q (up to 200 characters), id, topic (a thread ID), offset (nonnegative integer), and limit (1–50, default 20). Lists return total, offset, limit, items and nextOffset; null ends pagination. Unknown IDs return 404; malformed queries return 400. Other methods are unsupported. Results include public discoveries, source reports and relevant guide resources; staging dossiers, credentials and device-local reading history are excluded. Guide-aware retrieval helps broad questions reach a primary explanatory guide while preserving the existing response fields and stable-ID behavior.

Guide and model responses preserve source URLs or evidence IDs, provenance context, source publication dates, Bright publication or revision dates when known, and last-reviewed dates where available. A source check is not independent replication. Unknown dates remain unknown, and no response promises model or hardware compatibility.

Topic feeds use /feed.xml?topic=water or /feed.json?topic=water. Unknown topics return 404. Feeds include published discoveries with known Bright publication dates, not unreviewed signals.

HTTP and operational behavior

Public APIs support cross-origin GET, HEAD and OPTIONS without credentials. Responses include X-Agent-Contract-Version and X-Request-ID. Existing JSON error strings are preserved with X-Error-Code headers. Invalid inputs return 400; unknown IDs return 404; unexpected failures return JSON 500. Documentation must revalidate before reuse; schema, llms.txt and Markdown support ETags.

Bright logs bounded operational categories, status, result counts and handler duration, not raw questions or visitor identifiers. Hosting infrastructure may separately retain URLs and network data. See privacy. Scheduled external checks test the public contract every six hours.

Dates and authorship

Source publication is the historical source date. Bright publication records catalog publication, not a new breakthrough. Substantive updates come from recorded revisions or corrections; a source check alone is not an update. Unknown dates and unnamed reviewers stay unknown. AI-assisted editorial review is identified; it is not expert review or independent replication. Date-only records are serialized at UTC midnight in feeds as a format convention, not an observed publication time.

Experimental browser tools

Where the browser implements WebMCP, Bright offers bright_search, bright_evidence, bright_recent_updates, bright_guide, bright_latest_weekly and bright_check_claim. Ordinary links and HTTP remain available everywhere. This experiment does not create global discovery or establish compatibility with an untested agent. llms.txt is a community convention, not a promise of indexing or citations.