Two APIs, one key

Agent lifecycle: Create an agent with a persona, model preference, and lifecycle type. Persistent agents maintain state forever. Ephemeral agents are created per-request and destroyed after. On-demand agents spin up when needed and idle down. Scheduled agents run on cron triggers. Every agent gets isolated memory, knowledge, and tool access.

V4 enhancements: Durable-chat ensures agent conversations survive server restarts. Structured-chat enforces JSON schema output. Workflow-chat binds an agent to a specific DAG. Session-status and session/history give full visibility into running conversations.

V5 spawn: Spawn multiple sub-agents in parallel for fan-out tasks. Track spawn status, collect results, and merge back into the parent agent context.

Core memory: The agent persistent personality — facts, preferences, and instructions always in context. Append, replace, or delete core blocks. Every chat request automatically includes core memory.

Recall memory: Recent conversation turns stored with timestamps. Search semantically or retrieve the N most recent messages. Auto-pruned when context gets too large.

Archival memory: Long-term storage for facts extracted from conversations. Semantic search over all archived knowledge via Qdrant vectors (384-dim FastEmbed). Auto-consolidation promotes important recall memories to archival.

Temporal search: Query memories by time range — "what did we discuss last Tuesday?" Context building assembles the optimal context window from core + recall + archival automatically.

Creating a knowledge base: POST /v1/knowledge-bases with a name and optional agent_id. Upload documents — each is auto-chunked (default 512 tokens, 50 overlap) and embedded using FastEmbed bge-small-en-v1.5 (384-dim). Model-agnostic — same vectors work with any LLM.

Querying: POST /v1/knowledge-bases/:id/query with a search string returns the most relevant chunks ranked by cosine similarity. The intelligent tier automatically injects relevant KB chunks into agent context.

External knowledge (v5): The knowledge hub searches Anna's Archive, OpenAlex, Internet Archive, Google Books, Wikipedia, Crossref, Semantic Scholar, and Open Library in real-time. POST /v5/ks/enrich enriches documents with extracted entities and summaries.

V1 workflows: Simple create/run/delete with trigger configuration. POST /v1/workflows with a DAG definition and trigger type (api, cron, webhook, event). Run on demand or on schedule. Full run history.

V2 persistent workflows: Restate-backed durable execution. Workflows survive server crashes and restarts. Resume crashed runs via /v2/workflows/runs/:id/resume. Full run history and status tracking.

V4 batch processing: Submit batch jobs for high-volume parallel processing. Dead letter queue (DLQ) at /v4/dlq captures failed items for inspection and retry with /v4/dlq/:id/retry.

V5 named workflows: Named, reusable workflow definitions with run tracking. /v5/workflows/:name/run executes by name. /v5/workflows/:name/runs lists all past executions.

Registering MCP servers: POST /v1/mcp-servers with a name, transport type, and URL/command. The server is scanned for available tools and resources automatically.

V4 MCP tools: /v4/mcp/tools lists all available tools across connected MCP servers. /v4/mcp/tools/register registers a new tool from an MCP server for agent use.

V5 MCP hub: Browse 9,973+ community MCP servers by category. Search by name or capability. Ingest community servers into your account. Full stats on available tools per server. Register custom MCP servers for your agents.

7 guardrail layers: (1) Input Filtering — blocks prompt injections, jailbreaks, and adversarial inputs before they reach the model. (2) Output Validation — scans completions for policy violations, harmful content, and PII leaks. (3) PII Redaction — detects and masks personal data in both directions. (4) Topic Restriction — confines agents to approved topics, blocking off-topic queries. (5) Toxicity Detection — filters harmful, offensive, or inappropriate content. (6) Injection Defence — prevents tool output hijacking and indirect prompt injection through tool results. (7) Custom Rules — define your own allow/deny patterns with regex or keyword matching.

Per-agent configuration: Each agent can have its own guardrail rules. The /v5/guardrails/check endpoint validates content against all enabled layers. Full audit log at /v5/guardrails/logs records every check for compliance.

Auto-build: POST /v5/kg/auto-build processes conversation history and documents to automatically extract entities (people, organisations, concepts, locations) and their relationships. The graph grows organically as your agents have more conversations.

Entity & relation extraction: POST /v5/kg/entities adds individual entities with metadata. POST /v5/kg/relations creates typed relationships between entities. Both support custom entity types and relation kinds.

Graph query: POST /v5/kg/search performs semantic + structural queries across the graph. Returns subgraphs of related entities ranked by relevance. This goes beyond simple vector similarity — it follows relationship paths to find contextually relevant information that pure embedding search would miss.

V1 skills: Create, list, and delete custom skills. Each skill bundles a system prompt, tool configurations, and workflow hooks into a reusable template that any agent can activate.

V5 skills hub: Browse 6,000+ community skills organised into packs. POST /v5/skills/for-prompt auto-discovers relevant skills based on the current conversation intent. POST /v5/skills/fetch downloads and installs a skill by ID. GET /v5/skills/packs lists curated skill packs for common use cases (coding, research, writing, data analysis).

V1 tools: Create, list, and delete custom tools with JSON Schema input/output definitions. Tools are callable by any agent that has access.

V4 WASM tools: Sandboxed tool execution via WebAssembly. POST /v4/wasm/tools compiles and deploys a WASM module as a tool. POST /v4/wasm/tools/:id/invoke executes it in an isolated runtime — no network access, no filesystem, pure computation.

V5 auto-generate: POST /v5/tools/generate creates a tool from a natural language description. POST /v5/tools/generate-from-url scrapes an API documentation page and auto-generates a tool wrapper. POST /v5/tools/prefilter selects the most relevant tools for a given prompt.

Creating schedules: POST /v5/schedule with a name, cron expression, and target (agent or workflow). The scheduler uses tick-based execution — every minute, the system checks for due schedules and fires them.

Run tracking: GET /v5/schedule/runs shows the full history of schedule executions including success/failure status, timestamps, and output. GET /v5/schedule/:name shows the schedule config and next run time.

Management: DELETE /v5/schedule/:name removes a schedule. POST /v5/schedule/:name/cancel stops a running schedule. POST /v5/schedule/tick forces an immediate tick for testing.

A2A protocol: POST /v2/a2a/broadcast sends a task to all agents with a matching capability. GET /v2/a2a/discover finds agents by capability or skill. POST /v2/a2a/tasks creates a task and assigns it to a discovered agent. Agents can communicate back and forth via /v2/a2a/tasks/:id/send.

Agent cards: GET /v2/agents/:id/card returns the agent A2A card — a machine-readable description of the agent capabilities, skills, and available tools. PUT /v2/agents/:id/card updates it. Other agents use cards to decide who to delegate tasks to.

Spawning: POST /v2/spawn creates multiple sub-agents in parallel. GET /v2/spawn/:id tracks spawn status. Sub-agents are ephemeral — they complete their task and return results to the parent.

Auto-build: POST /v2/improve/build takes a description of what the agent needs and generates a tool, skill, or workflow to fulfil that need. The generated artifact is immediately available for the agent to use in subsequent requests.

Capability search: POST /v2/improve/search finds existing tools, skills, and APIs that match the agent current need. If no match exists, it triggers auto-build. This creates a self-reinforcing loop — agents identify gaps, search for solutions, and create what they need.

GET /v2/status shows the current v2 engine status including active improvements and generated artifacts.

Traces: GET /v5/observability/traces returns detailed request traces including model used, tokens in/out, latency, cache hit, and cost. Filter by agent, model, or time range.

Errors: GET /v5/observability/errors lists all errors with stack traces, request context, and retry status. Critical for debugging agent failures.

Metrics: GET /v5/observability/metrics returns aggregated metrics — requests per minute, average latency, token throughput, error rate, cache hit rate.

Snapshots: POST /v5/observability/snapshot captures a point-in-time snapshot of all system state. GET /v5/observability/snapshots lists historical snapshots for comparison.

Database: GET /v5/observability/database shows connection pool status, query performance, and storage metrics for the underlying data stores.

Queue-first architecture: When all concurrency slots are occupied, the system queues the request and waits for a slot to free up. Clients never receive 429 Too Many Requests — requests are always processed, just potentially delayed.

Configuration: POST /v5/rate-limits/configure sets rate limits per API key. GET /v5/rate-limits/config/:key shows current limits. GET /v5/rate-limits/check shows remaining quota. GET /v5/rate-limits/tiers shows the default limits per subscription tier.

Cleanup: POST /v5/rate-limits/cleanup removes expired rate limit entries. POST /v5/rate-limits/increment manually increments a counter (used internally by the proxy).

Event store: POST /v5/events appends an event to the log. GET /v5/events queries events by entity type, time range, or event kind. Every agent action, memory change, tool call, and workflow step is recorded as an event.

Replay: GET /v5/events/replay/:type/:id replays all events for a specific entity (agent, workflow, memory) in order, reconstructing its full state history. Essential for debugging and compliance audits.

Subscriptions: GET /v5/events/subscribe/:type opens a real-time SSE stream of events matching the filter. Build dashboards, trigger webhooks, or feed downstream systems.

Management: GET /v5/events/stats shows event counts and storage usage. POST /v5/events/trim removes events older than a specified age to manage storage.

Creating pipelines: POST /v5/pipelines with a name and a list of steps. Each step references a v5 endpoint (e.g., extract facts, check guardrails, store memory). Steps execute in order with output piped to the next step.

Execution: POST /v5/pipelines/:name/run executes the pipeline by name. GET /v5/pipelines/:name/runs shows all past executions with timestamps, duration, and success/failure status.

Common patterns: Ingest pipeline (upload document, extract facts, build KG, store memory). Guard pipeline (input check, model call, output check, PII scan). Memory pipeline (extract from conversation, consolidate tiers, promote to archival).

Discovery: GET /v5/plugins/categories lists all plugin categories. POST /v5/plugins/for-task finds plugins relevant to a specific task description. POST /v5/plugins/search does full-text search across the plugin catalog.

Installation: POST /v5/plugins/ingest imports a community plugin. POST /v5/plugins/register creates a custom plugin. POST /v5/plugins/install/:id installs it for agent use.

Execution: POST /v5/plugins/invoke/:id runs a plugin with input data. GET /v5/plugins/list shows all installed plugins. GET /v5/plugins/stats shows usage statistics.

Tree-of-thought: POST /v5/tot/quick explores multiple reasoning paths quickly (3-5 branches, shallow depth). POST /v5/tot/reason does deep analysis (10+ branches, multiple rounds of evaluation). Both return the best path with confidence scores.

Fact extraction: POST /v5/facts/extract pulls structured facts from unstructured text. POST /v5/facts/batch-extract processes multiple texts at once. POST /v5/facts/extract-and-store extracts facts and stores them in archival memory in one call. POST /v5/facts/extract-turn extracts facts from a single conversation turn.

Intent classification: POST /v5/intent/classify categorises user intent (question, command, creative, analytical) to route to the optimal processing path.

Summarization: POST /v5/summarize/agent/:id summarizes an agent full conversation history. POST /v5/summarize/messages summarizes a specific message list.

Registration: POST /v1/auth/register creates an account with email, password, name, and tier. Returns a JWT session token and an API key immediately. No email verification required — start using the API right away.

Login: POST /v1/auth/login returns a JWT session token (valid 30 days) and the user API key. The session token is used for dashboard operations. The API key is used for raw.designat.ing and api.designat.ing requests.

API keys: POST /v1/auth/api-keys generates additional keys with scoped permissions (e.g., read-only, agents-only). GET /v1/auth/api-keys lists all keys. DELETE /v1/auth/api-keys/:id revokes a key immediately.

Current user: GET /v1/auth/me returns the authenticated user details including tier, tokens used, and concurrency slots.

Plans: Basic (£79/mo, 20B tokens, 4 slots), Plus (£179/mo, 50B tokens, 4 slots), Pro (£299/mo, 100B tokens, 6 slots), Pro+ (£479/mo, 250B tokens, 10 slots), Ultra (£679/mo, 500B tokens, 20 slots), Enterprise (£999/mo, 1T tokens, 25 slots).

Checkout: POST /v1/billing/checkout generates a Square checkout URL. The user pays on Square hosted page, then the webhook at /v1/billing/webhook activates the plan automatically.

Subscription management: GET /v1/subscription/me shows current plan, tokens used, and renewal date. GET /v1/subscription/tiers lists all available tiers. POST /v1/billing/activate activates a plan manually. PATCH /v1/clients/:id/tier changes a client tier (admin).

Full catalog: GET /v1/models returns all available models with id, context_length, pricing, tool_use capability, model_class, and owned_by. GET /v1/models/all returns the full Featherless catalog (paginated).

Categorised catalog: GET /v1/catalog/models returns models grouped by category (flagship, premium, general, fast, economy). GET /v1/catalog/models/featured returns hand-picked best models per category. GET /v1/catalog/models/categories lists all available categories.

Search: GET /v1/models/search performs full-text search across model names, descriptions, and categories.

V4 capabilities: GET /v4/models returns models with enhanced capability metadata including tool_use, streaming, and vision support. GET /v4/models/capabilities groups models by capability.

Streaming: POST /v4/chat/completions supports SSE streaming with the standard OpenAI format. Tokens are delivered in real-time as they are generated. POST /v4/embeddings supports batch embedding requests.

Task detection: POST /v4/chat/detect-task analyses the user message and returns the detected task type (coding, reasoning, creative, factual) and recommended model tier. Use this to auto-route requests to the optimal model.

Batch: POST /v4/batch submits a batch of requests for asynchronous processing. Check status via the batch ID. Useful for high-volume fine-tuning data generation, bulk embeddings, or mass document processing.

Durable execution: POST /v4/agents/:id/durable-chat starts a conversation that survives server restarts. GET /v4/agents/:id/session-status checks the current state. GET /v4/agents/:id/session/history returns the full conversation history.

Templates: POST /v5/templates creates a reusable agent template with persona, model, and tool configuration. GET /v5/templates lists templates. POST /v5/templates/:name/render renders a template with variables. GET /v5/templates/:name/variant resolves the best variant for A/B testing. GET /v5/templates/analytics shows usage stats.

Episodes: POST /v5/episodes/ingest records a conversation episode with metadata. POST /v5/episodes/batch ingests multiple episodes. POST /v5/episodes/entity retrieves all episodes for an entity. POST /v5/episodes/temporal-facts returns time-annotated facts from episodes.

Skills hub: GET /v5/hubs/skills/search searches 6,000+ community skills. GET /v5/hubs/skills/top shows trending skills. GET /v5/hubs/skills/sources lists skill repositories. GET /v5/hubs/skills/fetch/:id downloads a skill.

Tools hub: GET /v5/hubs/tools/search searches 2,179+ public APIs. GET /v5/hubs/tools/categories categorises APIs by domain. POST /v5/apis/for-task recommends APIs for a specific task.

Workflows hub: GET /v5/hubs/workflows/search searches n8n workflow templates. GET /v5/hubs/workflows/live shows live workflow templates. GET /v5/hubs/workflows/categories categorises workflows.

Knowledge hub: GET /v5/hubs/knowledge/search cross-source search across 8 external knowledge sources.