For AI agents: a documentation index is available at https://docs.plungeai.com/llms.txt. Append .md to any page URL, or send Accept: text/markdown, to get markdown. Setup instructions for agents are at https://docs.plungeai.com/agents.md. One ozk_ key opens every plane, models included.

Documentation Index: fetch the complete documentation index at /llms.txt. Use this file to discover all available pages before exploring further.

plungeai-api-setup

Route map — the planes that are not bots, one line per operation

The One API 2.3.0 has 114 operations.

The One API 2.3.0 has 114 operations. Bots, executions, schedules, triggers, connections, memory, skills, templates and the Agent Call door are in bots-and-door.md; this file holds the rest, so every operation is named in this skill. Every route takes Authorization: Bearer ozk_YOUR_KEY except the three marked "no auth". Statuses below are the ones the spec lists per operation; 401 unauthorized is on every authenticated route. Regenerate a client from GET /v1/openapi.json rather than copying this.

Service

RouteDoes
GET /health (no auth)liveness {status, service, version}
GET /v1/openapi.json (no auth)this contract, OpenAPI 3.1
GET /v1/whoami{user_id, via: key|internal, tier, key_label}

Discovery (plungeai-discovery)

RouteDoes
GET /v1/discovery/searchquery q, kind (agents, skills, digital-twins, personas, experts, backgrounds, workflows, models, providers, connectors, plugins, mcp-servers), type, category, status (default active), tier, limit, offset, mode (hybrid, keyword, vector), fields (list, summary, full) → {cards, count, total, searchMethod}
POST /v1/discovery/recommend{type, task} → {recommended, card, score}; 400 invalid_json / invalid_request, 429
POST /v1/discovery/resolve{task, kinds?, category?, status?} → {outcome: match|uncertain|none, tool, operation, call: {required_params, example_cnl}, confidence, alternatives, mode: shadow|act}; 400 invalid_json
GET /v1/discovery/cards/{type}/{id}the card as markdown; type is singular (agent, skill, digital_twin, expert, background, workflow, model, provider, connector, plugin, mcp_server); 400 invalid_request, 404 not_found

Tools (plungeai-tools-connectors)

RouteDoes
GET /v1/toolsactive tool agents, limit ≤ 100 (default 50), offset → {tools, count}
GET /v1/tools/{id}the contract {agent_id, credential, name, description, operations, inputSchema, examples, output_type}; 404 unknown_tool
POST /v1/tools/{id}/execute{operation, params, prompt, format} → {ok, content, outcome, request_id}; 403 refused, 409 approval_required or duplicate_execution_id, 413, 422 invalid_params (echoes the contract), 424 (with connect), 429, 502 execution_failed, 503 agent_unavailable

Agents (plungeai-agents)

RouteDoes
GET /v1/agentsactive agents, limit ≤ 100 (default 50), offset → {agents, count}
GET /v1/agents/categories{categories, count}
POST /v1/agents/{id}/executeprompt (or input), persona, provider, model, maxTokens or max_tokens, temperature, top_p, reasoning_effort, thinking_level, system, messages, sync (default true), stream (default false: OpenAI-shaped SSE chunks, then data: [DONE]), user_context, format. 200 {content, workflow_id, task_id, request_id}; sync: false is 202 {workflow_id, task_id, request_id}. Errors: 400 missing_prompt, 403 agent_not_active, 404 unknown_agent, 409 duplicate_execution_id, 413, 422 invalid_params / unknown_model / empty_completion (raise max_tokens to 64 or more), 424, 429, 502 engine_error / result_unavailable, 503 agent_unavailable
GET /v1/agents/results/{workflowId}/{taskId}redeem an async result {content, content_type, workflow_id, task_id}; 404 not_ready (poll, 2 s or more), 409 execution_failed, 422 empty_completion

Workflow execution (plungeai-workflows → references/api.md)

POST /v1/workflows/execute (inline YAML or JSON), POST /v1/workflows/{id}/execute (saved), POST /v1/workflows/execute-stream and POST /v1/workflows/{id}/execute-stream (SSE), GET /v1/workflows/results/{workflowId}/{taskId} (redeem), POST /v1/workflows/executions/{id}/cancel (the same operation as POST /v1/executions/{id}/cancel). Bodies, events and errors are in that file; the saved-workflow CRUD and versions are in bots-and-door.md.

MCP pass-through and outbound runs (plungeai-mcp-setup, plungeai-results-traces)

RouteDoes
GET /v1/mcp/toolsthe platform's MCP tool list (JSON-RPC tools/list result)
POST /v1/mcpStreamable-HTTP MCP pass-through: any MCP client with an ozk_ key; JSON or SSE; a body that is not JSON is a JSON-RPC -32700 error, not the router envelope
POST /v1/mcp/runs{server_ids} (non-empty string array) → 201 {run_id, tools, connected, failed, warnings}; 502 no_servers_connected
GET /v1/mcp/runs/{id}/tools{tools, count}; 404 run_not_found once the run is closed or 30 minutes idle: open a new one
POST /v1/mcp/runs/{id}/call{name, args?, timeout_ms?} → {content}; 502 tool_call_failed
DELETE /v1/mcp/runs/{id}close the run (idempotent) → {success}

Traces

RouteDoes
GET /v1/traces/{id}{trace_id, spans, gateway_requests} for the x-trace-id you sent; 404 not_found for an unknown id or another key's (spans land asynchronously: retry in a few seconds)

Models plane (plungeai-models)

Same ozk_ key, Authorization: Bearer only.

RouteDoes
POST /v1/chat/completionsmodel, models[] (ordered fallback), sort (price, latency, throughput), messages, stream, temperature, max_tokens, top_p; @preset/<slug> is a model name. 402 byok_required, 403 model_not_allowed / content_blocked, 404 preset_not_found / model_not_found, 405 method_not_allowed, 429 rate_limit_exceeded / spend_cap_exceeded, 503 money_plane_unavailable
POST /v1/embeddings{model, input}
POST /v1/completionslegacy {model, prompt (one string), max_tokens?, stream?}, answered as text_completion
GET /v1/modelsthe priced catalog
GET /v1/models/{author}/{slug}/endpointsthe endpoint that serves one model, with per-token pricing
GET /v1/generation?id=usage and cost of one completion (gen-… id or x-request-id)
GET /v1/credits{data: {total_credits, total_usage}} in USD
GET /v1/keythe calling key: label, monthly limit, usage, rate limit

The OpenRouter drop-in base /api/v1 serves the same eight operations (POST /api/v1/chat/completions, /embeddings, /completions; GET /api/v1/models, /generation, /credits, /key, /models/{author}/{slug}/endpoints) with OpenRouter-shaped errors ({"error":{"code": <number>, "message", "metadata"}}). Point an OpenRouter client's base URL there; new code uses /v1.

Planned: TI-33

Search is not available yet. Until it ships, use the page index or browse the sidebar.

Planned: TI-34

The docs assistant is not available yet. You can hand these docs to your own assistant instead.