Just launched: AI Lead Finder

Changelog

New features, improvements, and fixes shipped to Coherence. For what's coming next, see the roadmap.

Atom feedJSON
API

Official TypeScript SDK

@coherenceos/sdk is now on npm — a fully typed, zero-dependency TypeScript client covering every /v1 endpoint, including SSE agent streaming.

  • npm install @coherenceos/sdk — the official TypeScript / JavaScript client is on npm. Fully typed, zero runtime dependencies, ESM + CommonJS, MIT licensed, Node 18+.
  • Covers every /v1 endpoint: modules, records (including bulk operations), references, cross-module search, activity feeds, and agents — with agents.stream() exposing the SSE agent run as a typed for await loop.
  • The SDK's operation set is contract-tested against the canonical OpenAPI spec, so the client can never silently drift from the API. See the TypeScript SDK reference.
API

Machine-readable error codes, complete live OpenAPI spec, versioning policy

Every API error now carries a stable machine-readable code, the live OpenAPI spec covers the full endpoint surface, and we published our versioning & deprecation policy.

  • Every error response now includes a machine-readable code field (e.g. rate_limit_exceeded, validation_error, not_found) alongside message and statusCode. The catalog is additive-only — codes are never renamed or removed. See Errors for the full catalog.
  • The live GET /v1/openapi.json endpoint now serves the complete spec covering every public endpoint, generated from a single canonical contract shared with the API implementation.
  • We published our versioning & deprecation policy: breaking changes only ship under a new version prefix, and deprecated behavior gets at least 90 days notice, announced in this changelog.
API

API reference docs accuracy overhaul

The API reference documentation now describes only shipped behavior, with planned capabilities clearly labeled.

  • Documented endpoints, parameters, and response shapes are verified against the live /v1 implementation.
  • Capabilities that are planned but not yet shipped are explicitly labeled as such instead of being documented as available.
  • The published OpenAPI spec and the reference docs are kept in sync with the deployed API surface.
Agents

Streaming responses for the Agents API

POST /v1/agents/messages now streams over Server-Sent Events, so agents and UIs can render progress as the run happens instead of waiting for the full response.

  • POST /v1/agents/messages now streams over Server-Sent Events — opt in with an Accept: text/event-stream header or a ?stream=true query parameter, using the same request body and agents:write scope.
  • The stream emits named open, token, tool_call, done, and error events so agents and UIs can render the answer token-by-token and surface tool activity instead of blocking until the run finishes. Raw model reasoning and raw tool output are never streamed, matching the privacy of the synchronous response.
  • The synchronous JSON response remains the default when neither the header nor the query parameter is present; the done event carries the exact same body.
API

API v1: filtering, bulk operations, search, activity feeds, and rate limit headers

The public API expanded with record filtering, bulk endpoints, module schema reads, cross-module search, read-only activity feeds, and standard rate limit headers.

  • Record listing now supports filter[field] params, a JSON advancedFilter, label and type filters, and field selection.
  • New bulk endpoints: POST /v1/modules/{moduleSlug}/records/bulk create and bulk delete.
  • Full module schema reads: GET /v1/modules/{moduleSlug} (fields, views, references) and GET /v1/modules/{moduleSlug}/views.
  • Cross-module search via POST /v1/search, with results filtered to modules the API key's role can view.
  • Read-only activity feeds: GET /v1/activity, plus per-module and per-record variants.
  • Every /v1 response now carries X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset; 429 responses include Retry-After.
Agents

Agent discovery artifacts

getcoherence.io now publishes machine-readable discovery artifacts so AI agents can find and integrate with Coherence on their own.

  • llms.txt and llms-full.txt for LLM crawlers, plus markdown views of key pages (/index.md, and content negotiation via Accept: text/markdown).
  • agents.md — an integration guide written for agents, covering authentication, the endpoint surface, and when to delegate to Nash.
  • OpenAPI 3.1 spec published at getcoherence.io/openapi.json.
  • .well-known/ cards: agent-card.json (A2A), agent-skills, and mcp/server-card.json.
Agents

MCP server published to npm

The official Coherence MCP server, @coherenceos/mcp-server, is now on npm so MCP clients can work with your workspace.

  • Run with npx @coherenceos/mcp-server and authenticate with a Coherence API key — works with Claude, Cursor, Cline, and any other MCP client.
  • Read and write workspace data (modules, fields, records) through the public API.
  • The chat_with_agent tool delegates multi-step tasks to Nash, the workspace agent, governed by your workspace's approval rules.
API

Public REST API v1

Coherence now has a public REST API at api.getcoherence.io/v1, authenticated with scoped API keys.

  • Records CRUD on any module — list, get, create, update, and soft-delete via /v1/modules/{moduleSlug}/records.
  • Module and field discovery: GET /v1/modules and GET /v1/modules/{moduleSlug}/fields.
  • Scoped sk_live_ API keys with per-endpoint scope enforcement (records:read, records:write, workspace:read, and more), managed from Settings → API Keys.
  • Machine-readable OpenAPI spec served at /v1/openapi.json.