Headless & Agent Access

Surfaces

Five ways to reach the same platform.

Every surface works with the same domain objects: projects, questions, answers, reviews, sources and jobs. Same permissions. Same review rules.

REST API

Full JSON access to projects, answers, reviews and jobs, with workspace permissions enforced.

POST /v1/answers

CLI

Script imports, exports and operations from a terminal using the same access model as the API.

cognaire respond import …

MCP Tools

Let compatible AI agents call Respond tools without bypassing permissions or review thresholds.

tool: cognaire.respond.*

Email Intake

Forward a CV, RFP or questionnaire to a project address and keep routing rules intact.

acme-2026@in.cognaire.com

SDK

Typed clients for product teams that need controlled access to Respond workflows.

@cognaire/respond-sdk

In practice

Same request, four surfaces.

Answering a single security question from your approved content. Same grounding, permissions and review rules - only the entry point changes.

# Ask one grounded question from the terminal
$ cognaire respond answer \
    --project acme-2026 \
    --question "Describe your data retention policy." \
    --grounding policy:security,prior:caiq-2025

→ answer      "Customer data is retained for 90 days…"
→ confidence  0.91
→ sources     policy:security§3.2, prior:caiq-2025#q44
→ review      pending  (SME: priya@cognaire.com)
→ job         8f2a-91c3 # audit-logged

POST /v1/answers  HTTP/1.1
Host:   api.cognaire.com
Auth:   Bearer sk_live_…

{
  "project":    "acme-2026",
  "question":   "Describe your data retention policy.",
  "grounding":  ["policy:security", "prior:caiq-2025"]
}

// 200 OK
{
  "answer":     "Customer data is retained for 90 days…",
  "confidence": 0.91,
  "sources":    [
    { "ref": "policy:security", "loc": "§3.2" },
    { "ref": "prior:caiq-2025", "loc": "#q44" }
  ],
  "review":     { "status": "pending", "sme": "priya@cognaire.com" },
  "job_id":     "8f2a-91c3"
}

import { Cognaire } from "@cognaire/respond-sdk";

const respond = new Cognaire({ apiKey: process.env.COGNAIRE_KEY });

const result = await respond.answers.create({
  project:   "acme-2026",
  question:  "Describe your data retention policy.",
  grounding: ["policy:security", "prior:caiq-2025"],
});

if (result.confidence < 0.8) {
  // Platform auto-routes low-confidence answers to SMEs.
  // Your code can also subscribe and react:
  await respond.webhooks.notify("answer.reviewed", result.id);
}

# An agent invokes Respond as an MCP tool
tool:    cognaire.respond.answer_question
args:
  project:   acme-2026
  question:  "Describe your data retention policy."
  grounding: [policy:security, prior:caiq-2025]

# Platform enforces the same governance as the UI:
→ permission check   role=agent:external-bid  ✓
→ retrieval scope    policy:security, prior:*  ✓
→ confidence         0.91  (threshold: 0.80)
→ audit logged       job 8f2a-91c3
→ response returned  with provenance + sources

# Agents are just another user — with scoped, auditable tools.

Principles

Three rules every surface follows.

The browser is not the control layer.

Permissions, retrieval scoping and review thresholds live at the platform layer. A UI click, CLI command and agent call all route through the same checks.

Source context travels with the answer.

Responses can include sources, confidence signals and job identifiers for downstream review. Teams can inspect how an answer was produced instead of treating it as a black box.

Human review remains configurable.

Low-confidence answers, sensitive topics and threshold breaches can be routed to reviewers even when the request starts from an API, email or agent workflow.

MCP tool surface

A partial view of Respond's agent tools.

Each tool is scoped by role and enforces the same permissions as the UI. Confidence thresholds, review routing and approval gates apply regardless of how the request arrives — so human judgement stays in the loop wherever it's configured to be.

Questionnaires & RFP

# cognaire.respond — questionnaire tools (excerpt) # Role: agent:bid-writer | agent:rfp-reviewer tool: cognaire.respond.list_projects List projects and documents visible to the caller. args: account_id?, status? tool: cognaire.respond.ingest_questionnaire Parse an RFP or questionnaire into structured question groups ready for answering. args: project_id, file (xlsx | csv | pdf | docx) tool: cognaire.respond.answer_question Answer a single question from the governed corpus, with grounding, confidence and source attribution. args: project_id, document_id, question, corpus verbosity?, model_tier? tool: cognaire.respond.bulk_answer Answer all open questions in a document against the governed corpus in a single operation. args: project_id, document_id, corpus model_tier?, verbosity? tool: cognaire.respond.search_corpus Retrieve grounded content from the knowledge base for a given query. args: corpus, query, top_k? tool: cognaire.respond.request_review Route an answer to a human SME by role or username. Keeps human judgement in the loop where it matters. args: project_document_id, item_key, reviewer? tool: cognaire.respond.export_document Export a completed questionnaire in the buyer's required format. args: project_id, document_id format (sap_ariba | coupa | csv | docx)

Recruitment & Talent

# cognaire.recruit — recruitment tools (excerpt) # Role: agent:recruiter | agent:bench-manager tool: cognaire.recruit.list_accounts List client accounts and active roles visible to the caller. args: filter?, owner? tool: cognaire.recruit.process_cv Process a CV through the platform pipeline and produce a structured profile and branded output. args: file (pdf | docx), job_brief? tool: cognaire.recruit.list_profiles Filter the talent pool by availability, seniority, location, work rights, clearance, and more. args: availability_weeks?, seniority_levels? location_cities?, work_rights_code? min_years_experience?, limit? tool: cognaire.recruit.search_profiles Search the talent pool against a role brief or free-text query. args: project_id, document_id, query? scope? (pool | bench), top_k? tool: cognaire.recruit.analyse_fit Assess one or more profiles against a role brief. A consultant reviews and signs off before the output leaves the platform. args: project_id, document_id, talent_profile_ids[] tool: cognaire.recruit.suggest_skills Suggest skills and occupation codes from a role description — useful for enriching search briefs. args: description, title? tool: cognaire.recruit.pack_review Compile a candidate shortlist into a client-ready submission pack, routed for consultant review. args: project_id, document_id, talent_profile_ids[]

Build on the same foundation your users already trust.

Request API, CLI or MCP access with your Cognaire workspace.

Book a Demo Platform overview

Beyond the browser. For teams and agents.