Managed Agents API Coming soon

Lojiq's agents are the product: the sales caller that works a list, the receptionist that answers your line, the texting journeys that keep a thread alive, the follow-up agent that never forgets a callback. The Managed Agents API lets you run those exact agents inside your own software, with Lojiq operating them, and pay only for what they do.

What "managed" means#

Today's Voice agents endpoints let you bring your own prompt and voice and place calls with it — you write the agent, Lojiq runs the call. A managed agent is the other way round: Lojiq brings the agent — its playbook, objection handling, compliance behaviour, memory of each lead, retries and follow-up logic, the same agents that ran more than 110,000 calls and 54,000 text enrollments in the last 30 days — and you bring the business facts and the lead. You get the outcome, not a prompt to maintain.

The agents you will be able to run#

How an integration would look#

Three calls, then listen.

1. Look at the catalog#

   GET /managed-agents/catalog lists the agent types enabled for your organization, with their version, channels, what each needs from you, the pricing unit and the guardrails it runs under. Never the prompt.

2. Create your copy and turn it on#

   POST /managed-agents with the type, a name and your config (validated against the type's schema). Attach a number with POST /managed-agents/{id}/phone-number, then POST /managed-agents/{id}/activate, which runs the launch checks (balance, number present, A2P registration for texting, compliance preset complete) and answers 409 activation_blocked with blocked_reasons if anything is missing.

3. Give it work#

   POST /managed-agents/{id}/jobs with a lead (inline, with the same consent fields as POST /leads, or by lead_id) or a task ("call this person Tuesday 3 pm about the quote"), a token budget, and an Idempotency-Key. Receptionists create their own jobs: every inbound call is one.

   Draft — shapes may change before launchCopy

   curl -s -X POST "https://api.lojiq.ai/v1/managed-agents/8c1d…/jobs" \
     -H "Authorization: Bearer $LOJIQ_KEY" -H "Content-Type: application/json" \
     -H "Idempotency-Key: 4e5f2a0c-7b1d-4c8e-9a2f-3b6d1e0c5a77" \
     -d '{
       "kind": "work_lead",
       "lead": { "phone_number": "+15551234567", "first_name": "Avery",
                 "consent_source": "web_form", "consented_at": "2026-10-02T16:58:00Z" },
       "budget_tokens": 2000,
       "metadata": { "crm_id": "opp_8812" }
     }'
   

   202 Accepted (draft)Copy

   {
     "id": "9f0e1d2c-3b4a-4c5d-8e6f-7a8b9c0d1e2f",
     "agent_id": "8c1d…",
     "kind": "work_lead",
     "status": "queued",
     "lead_id": "4b1e0d1a-3c4e-4d0a-9f2b-7e1a2b3c4d5e",
     "budget_tokens": 2000,
     "spent_tokens": 0,
     "next_attempt_at": "2026-10-02T17:05:00Z",
     "created_at": "2026-10-02T17:03:11Z"
   }
   

4. Get the outcome#

   Subscribe to job.* with the webhooks you already have, or poll GET /jobs/{id}. A job ends with one outcome: appointment_booked, transferred, interested, not_interested, voicemail_left, do_not_call or no_answer (after the configured attempts), with the transcript and recording links of every attempt.

   job.completed — webhook body (draft)Copy

   {
     "id": "c1d2e3f4-a5b6-4c7d-8e9f-0a1b2c3d4e5f",
     "type": "job.completed",
     "organization_id": "7f6e5d4c-3b2a-4190-8877-665544332211",
     "occurred_at": "2026-10-02T17:21:40Z",
     "data": {
       "job_id": "9f0e1d2c-3b4a-4c5d-8e6f-7a8b9c0d1e2f",
       "agent_id": "8c1d…",
       "lead_id": "4b1e0d1a-3c4e-4d0a-9f2b-7e1a2b3c4d5e",
       "outcome": "appointment_booked",
       "appointment_id": "8d7c6b5a-4f3e-4d2c-9b1a-0f9e8d7c6b5a",
       "attempts": 2,
       "calls": ["5e6f7a8b-…", "6f7a8b9c-…"],
       "summary": "Agreed to a consultation on Thursday at 2pm.",
       "spent_tokens": 1180
     }
   }
   

What you pay#

API requests are free. A job bills exactly what the same work costs in the Lojiq app, through the same token wallet (1 token = $0.001), at the rates below — plus a per-job fee per agent type that will be published with the launch entry. Every job carries a token budget; when it is reached the job pauses with job.waiting (reason: budget_reached) rather than silently dialing on, and you raise the budget with PATCH /jobs/{id}. A daily cap per agent and a monthly cap per key sit on top.

Plans and packs are the ones the app sells; an API-only organization is simply an organization with keys and a wallet and no seats.

Rules the API cannot switch off#

These are already enforced for the app and will wrap every managed-agent attempt. The catalog lets you make them stricter, never looser.

• Consent. Outbound sales voice and SMS need recorded consent on the lead (consent_source and consented_at); a job without it is refused with 403 consent_required.
• Do-not-call. Suppression lists are checked at job creation (403 destination_on_dnc) and again before every attempt, failing closed. A lead who says "stop" on a call or texts STOP ends the job with outcome do_not_call.
• Calling hours. Voice 8:00–21:00 in the destination's local time; texting falls back to the most conservative US window when the time zone is unknown. Outside the window a job waits (job.waiting, reason: calling_window, with next_window_at).
• Frequency. One dial per lead per local day by default, per-number daily caps, per-journey hourly and daily text caps.
• AI disclosure and recording. The agent identifies itself as an AI assistant per Lojiq's disclosure policy; two-party-consent states get the recording disclosure or recording off.
• Hard stop. No AI call runs longer than 15 minutes.
• 10DLC. No texting agent activates without an approved A2P brand and campaign.
• People first. A lead owned by one of your human reps is not worked by an AI job (409 lead_owned_by_human); organization-level kill switches pause every job.

Sandbox#

In a sandbox organization, jobs run in dry-run by default: the agent's plan and a synthetic outcome are produced, job.* and call.* events fire with simulated: true, no carrier is touched and no tokens are spent. A job may dial a phone on the sandbox's allow-list for real (your own), billing the sandbox's test grant. Texting in a sandbox is always simulated.

Draft surface#

Illustrative; final paths will be published with the contract. Scopes: managed_agents:read, managed_agents:write, jobs:read, jobs:write, usage:read.

Events (draft)#

Delivered by the same signed dispatcher as every other event: managed_agent.activated, managed_agent.paused, job.created, job.started, job.attempt (one per dial or send, with the call_id or message_id), job.waiting (reason: calling_window, budget_reached, needs_human), job.needs_human (transfer requested, escalation phrase, emergency keyword), job.completed, job.failed, job.cancelled, candidate.interviewed (scores, never a decision). Existing call.*, appointment.* and lead.* events keep firing for the same activity and carry job_id.

Principles that will not change#

• Your organization is the key. Same API keys, same scopes model, same host, same webhooks.
• Budgets. Every job carries a budget; a job never spends past it without you raising it.
• Consent and compliance travel with the agent. Opt-outs, DNC, calling hours and frequency caps are enforced by Lojiq, not re-implemented by you — and cannot be disabled through the API.
• No prompts, no raw SMS. You never see or ship a prompt, and there is no endpoint that sends an arbitrary message to an arbitrary number.
• Idempotency. Every POST that can cost money takes an Idempotency-Key.
• Versioned. Managed Agents launches with its own changelog entry and joins the versioning policy at that point.

Tell us what you would build#

The order in which agents and channels ship is driven by real integrations. Write to support@lojiq.ai with what your software does and which agent you would run first, and we will put you on the early-access list.