Decision-first
Start with the user's choice, not an endpoint.
User decision
Choose an agent framework for Cloudflare Workers
Request: Choose an open-source agent framework that can run on Cloudflare Workers and explain the evidence.
Shortest REST pathGET /api/trust -> POST /api/workflow -> GET /api/project/cloudflare/agents
Shortest MCP pathget_agent_workflow -> get_project
Expected fieldsshortlist, decision_summary, classification, caveats, metadata.source
Example final answerFor this Cloudflare-specific target, cloudflare/agents is the current shortlist match with high recommendation confidence and matched deployment, category, and Cloudflare-readiness constraints. The project record says it is designed for Cloudflare-native, stateful agents. Caveat: classification and several quality signals are low-confidence or partial, so verify the deployment guide and run a Workers proof of concept before adoption.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionInspect the repository deployment instructions and compare a proof of concept against any non-Workers runtime requirement.
User decision
Compare two coding agents for local use
Request: Compare Codex and OpenCode for a local coding-agent workflow.
Shortest REST pathGET /api/trust -> POST /api/compare
Shortest MCP pathcompare_projects
Expected fieldscontext, decision_matrix, winner, reasoning, metadata.snapshot_id
Example final answerFor the local-deployment context in this snapshot, openai/codex ranks ahead of anomalyco/opencode. Both match the local preference. Caveat: both carry a 24/100 maintenance signal in the current comparison. Treat Codex as the contextual leader, not a universal winner, and inspect recent releases and issue activity for both before choosing.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionRun the same representative repository task in both clients and compare tool reliability, edit quality, and operational cost.
User decision
Check an unfamiliar dependency before adding it
Request: Should I add cloudflare/agents to a Python-first agent stack?
Shortest REST pathGET /api/trust -> GET /api/project/cloudflare/agents
Shortest MCP pathget_project
Expected fieldssummary.good_for, summary.not_good_for, classification, quality_signal_confidence, caveats
Example final answerDo not select cloudflare/agents by default for a Python-first stack. Its strongest fit is Cloudflare-native, stateful agents on Workers, while the project summary explicitly lists Python-first stacks as a poor fit. Caveat: the current record has incomplete classification and quality evidence, so confirm language and integration requirements in the upstream repository before ruling it in or out.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionConfirm whether the application can use a TypeScript Workers boundary; otherwise compare a Python-native framework.
User decision
Find a credible LangChain alternative
Request: Find maintained alternatives to LangChain and distinguish direct replacements from adjacent options.
Shortest REST pathGET /api/trust -> GET /api/alternatives/langchain -> POST /api/compare
Shortest MCP pathget_alternatives -> compare_projects
Expected fieldsalternative_matches, fit_summary, replacement_risk, caveats, metadata.source
Example final answersuperagent-ai/superagent is the top direct alternative in this snapshot, with 93/100 similarity and low replacement risk. TransformerOptimus/SuperAGI is another direct-category candidate. run-llama/llama_index is adjacent rather than drop-in because its category differs. Caveat: estimated activity signals and API compatibility still need upstream verification.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionCompare the top direct candidate with LangChain against the exact integrations, deployment model, and migration surface you use.
User decision
Choose a vector database for Docker
Request: Recommend a vector database that can be self-hosted with Docker.
Shortest REST pathGET /api/trust -> POST /api/recommend -> POST /api/compare
Shortest MCP pathrecommend_project -> compare_projects
Expected fieldsrecommendations, matched_constraints, risk_flags, confidence, caveats
Example final answerRedis and Milvus are tied at 62/100 in this Docker-constrained snapshot, with Qdrant close behind at 61/100; all are medium-confidence exploration candidates. Do not choose from score alone. Each carries a weak maintenance warning in the indexed evidence and is unsuitable for edge-only Workers deployment without adaptation.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionCompare operational footprint, filtering requirements, scale, backup model, and client-language support before running a workload benchmark.
User decision
Replace an archived Python agent framework
Request: Find a maintained replacement for the archived microsoft/TaskWeaver project for local Python tool execution.
Shortest REST pathGET /api/trust -> POST /api/recommend -> GET /api/project/langchain-ai/langchain -> POST /api/compare
Shortest MCP pathrecommend_project -> get_project -> compare_projects
Expected fieldsrecommendations, matched_constraints, confidence, classification, caveats, metadata.snapshot_id, external_evidence
Example final answerGitHub reports microsoft/TaskWeaver as archived. For a local Python agent-framework replacement, Git.Top ranks langchain-ai/langchain first at 65/100 with medium confidence, followed by pydantic/pydantic-ai at 45/100 with medium confidence. Caveat: this is not a drop-in compatibility claim; LangChain's indexed classification is low-confidence and the candidates need API, tool-execution, and migration testing against the TaskWeaver workload.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
External evidencegithub_api microsoft/TaskWeaver: archived=true, disabled=false (checked 2026-07-31)
Next actionList the TaskWeaver APIs and execution behaviors in use, then run the same tool task in LangChain and Pydantic AI before selecting a migration target.
User decision
Choose a local LLM runtime
Request: Choose among Ollama, llama.cpp, and GPT4All for a local inference workflow.
Shortest REST pathGET /api/trust -> GET /api/search -> POST /api/compare -> GET /api/project/ollama/ollama
Shortest MCP pathsearch_projects -> compare_projects -> get_project
Expected fieldsprojects, decision_matrix, winner, classification, quality_signal_confidence, caveats, metadata.snapshot_id
Example final answerFor the local-deployment context in this snapshot, ollama/ollama is the contextual winner over ggml-org/llama.cpp and nomic-ai/gpt4all. All three have a 79/100 Git.Top score and a 24/100 maintenance signal; llama.cpp has the strongest indexed category confidence. Caveat: the comparison does not model hardware, quantization, model-format, or throughput requirements, so Ollama is a workflow starting point rather than a universal runtime winner.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionBenchmark the same model and prompt set on target hardware, including memory use, startup time, throughput, model-format support, and API compatibility.
User decision
Choose an MCP server for GitHub automation
Request: Choose an MCP server for repository and pull-request automation on GitHub.
Shortest REST pathGET /api/trust -> GET /api/search -> GET /api/project/github/github-mcp-server
Shortest MCP pathsearch_projects -> get_project
Expected fieldsprojects, project_id, summary, classification, git_top_score, quality_score, caveats, metadata.source
Example final answergithub/github-mcp-server is the exact-purpose candidate in this snapshot: its indexed description is an MCP server for GitHub developer tools, its MCP-server classification is high-confidence, and its Git.Top score is 79/100. Caveat: its quality score is 43/100, maintenance signal is 24/100, and the generic recommender gives this exact use case low confidence. Prefer it for a proof of concept because of purpose fit, then validate authentication scopes and every mutation tool before operational use.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionConnect it with a least-privilege test token and verify read-only repository queries before testing issue or pull-request mutations in a disposable repository.
User decision
Assemble a RAG stack with observability
Request: Draft a self-hosted Docker RAG stack with observability and provide citable trust evidence.
Shortest REST pathGET /api/trust -> POST /api/grp/query -> GET /api/project/run-llama/llama_index -> GET /api/project/SigNoz/signoz
Shortest MCP pathget_trust_gate -> git_top_grp_query -> get_project
Expected fieldsdecision, checks, solution_paths, recommended_stack, stack_roles, tradeoffs, metadata.data_source, caveats
Example final answerUse the GRP result only as an architecture draft: LlamaIndex fills the memory role, LangChain the agent-core role, Docker the runtime, and SigNoz the observability role. The leading path scores 72/100. Do not present this as production-ready: the Trust Gate is block because sync and hot-corpus freshness fail, and the bounded graph reports missing or weak protocol, tool-execution, and LLM-access coverage.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionPin one project per required role, verify supported integrations and licenses upstream, then build a traced ingestion-and-retrieval proof of concept before re-running the trust gate.
User decision
Distinguish a collection from an executable project
Request: Can I adopt run-llama/awesome-rag as my RAG runtime, or should I use LlamaIndex?
Shortest REST pathGET /api/trust -> GET /api/project/run-llama/awesome-rag -> GET /api/project/run-llama/llama_index -> POST /api/compare
Shortest MCP pathget_project -> compare_projects
Expected fieldsproject_kind, collection_metadata, summary.install, decision_matrix, winner, caveats, metadata.source
Example final answerDo not adopt run-llama/awesome-rag as a runtime. Git.Top identifies it as a curated collection with an estimated 100 items and no direct install step. For an executable local RAG project, run-llama/llama_index wins the comparison and carries a 91/100 agent score and 88/100 quality score. Caveat: the collection remains useful for discovery, but its 80/100 Git.Top score is not runtime suitability and should not be compared as if it were an installable library.
Verificationd1 d1:504:2026-06-20T00:00:00Z, checked 2026-07-31; local D1 scope, not a production snapshot
Next actionUse the collection to expand the shortlist, then evaluate LlamaIndex and other executable candidates against ingestion, retrieval, storage, and deployment requirements.
Protocol examples
Copyable calls and response fields.
REST Example
Require D1-backed data
GET/api/health
Check whether production recommendations can be treated as high-confidence.
curl https://git.top/api/health
Inspectdb, metadata.source, sync_health, sync_freshness
Trust checksmetadata.source=d1, db=available
NextAdd require_d1=true to critical read APIs when fallback data should be rejected.
REST Example
Choose the right surface
GET/api/agent-map
Map a user intent to the matching human page, REST endpoint, MCP tool, output fields, and trust fields.
curl https://git.top/api/agent-map
Inspectsurfaces[].concept, surfaces[].rest, surfaces[].mcp_tools, surfaces[].trust_fields
Trust checkstrust_policy.high_confidence_source, surfaces[].trust_fields
NextUse the selected surface before calling project lookup, recommendations, graph, alternatives, compare, score, Atlas, or quality.
REST Example
Run a constrained selection workflow
GET/api/workflow
Choose a Cloudflare-ready agent framework with an explainable shortlist and trust policy.
curl "https://git.top/api/workflow?intent=choose%20a%20Cloudflare-ready%20agent%20framework&deployment=cloudflare&category=agent_framework&cloudflare_ready=true&limit=5&require_d1=true"
Inspectrecommended_sequence, shortlist, trend_context, trust_policy, metadata.source
Trust checksmetadata.source=d1, shortlist[].confidence, trust_policy.disclose_when
NextInspect the top project, graph context, score explanation, and compare matrix before answering.
REST Example
Recommend from structured JSON
POST/api/recommend
Use POST when an agent has nested constraints or wants a stable request body.
curl -X POST "https://git.top/api/recommend?require_d1=true" -H "content-type: application/json" -d '{"use_case":"build browser automation agents","constraints":{"deployment":"docker","category":"browser_agent"},"limit":5}'
Inspectrecommendations[].decision_summary, recommendations[].fit_profile, recommendations[].risk_flags, recommendations[].ranking_signals
Trust checksmetadata.source=d1, recommendations[].confidence, recommendations[].unmatched_constraints
NextReturn the top fit, one credible alternative, caveats, and the data source.
REST Example
Compare candidates
GET/api/compare
Turn a shortlist into a decision matrix instead of a star-only ranking.
curl "https://git.top/api/compare?repos=cloudflare/agents,langchain-ai/langchain,run-llama/llama_index&deployment=cloudflare&require_d1=true"
Inspectsummary, decision_matrix, winner, reasoning, next_actions
Trust checksmetadata.source=d1, context.deployment, decision_matrix[].tradeoffs
NextMove into graph or score endpoints when the decision matrix needs evidence.
REST Example
Move from ecosystem map to comparison path
GET/api/journeys
Use Atlas journeys to turn an ecosystem into ordered exploration steps and concrete compare URLs.
curl https://git.top/api/journeys?limit=8&require_d1=true
Inspectjourneys[].steps, comparison_paths, comparison_paths[].api_href, stats.comparison_path_count
Trust checksmetadata.source=d1, comparison_paths[].context, journeys[].ecosystem_id
NextRun the selected comparison_paths[].api_href, then inspect score and graph evidence.
MCP Example
Discover MCP tools
POST/mcp
Inspect the available MCP tools before calling a tool by name.
curl -X POST https://git.top/mcp -H "content-type: application/json" -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
Inspectresult.tools[].name, result.tools[].inputSchema
Trust checksUse require_d1 in tool arguments for strict production reads
NextPick the tool whose name matches the Agent Map concept.
MCP Example
Fetch project knowledge through MCP
POST/mcp
Use MCP when an agent runtime prefers JSON-RPC tools over direct REST calls.
curl -X POST https://git.top/mcp -H "content-type: application/json" -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"get_project","arguments":{"owner":"cloudflare","repo":"agents","require_d1":true}}}'
Inspectresult.project, result.knowledge, result.quality_signal_confidence, result.metadata.source
Trust checksresult.metadata.source=d1, classification evidence, quality_signal_confidence
NextUse compare_projects or get_project_graph after fetching a candidate.
GRP Example
Plan a project stack with GRP
POST/api/grp/query
Use graph reasoning when the user asks for a plan, stack, or project set instead of one repository.
curl -X POST "https://git.top/api/grp/query?require_d1=true" -H "content-type: application/json" -d '{"goal":"compose an autonomous coding stack with MCP tools","mode":"compose","constraints":{"agent_ready":true}}'
Inspectsolution_paths, recommended_stack, nodes, edges, metadata.data_source
Trust checksmetadata.data_source.source=d1, nodes[].repo, confidence
NextValidate each recommended project with /api/project, /api/score, and /api/compare.