Conduir

Agents

Your agents — directory, connections, and setup guide.

New agent
CCortex MaaSMemory as a ServiceConnect an agent
Agent memory connection

Give the agent you already run durable memory

Keep your agent runtime, model provider, instructions, tools, and optional knowledge workflow. Add Cortex MaaS to recall relevant context before work and remember approved outcomes afterward.

Bring your own agent

Connect an agent running inside or outside your Conduir environment.

Choose the boundary

Use private agent memory or a shared memory already approved for your organisation.

Least access

Grant recall only, or recall and remember, according to the agent's role.

Current evaluation address

Trust the evaluation certificate before setup

The current bare-IP evaluation uses a private trust certificate. Download the public certificate, compare its SHA-256 fingerprint with your approved onboarding contact, and transfer it to the machine running the agent. Never disable certificate verification.

Download evaluation trust certificate
SHA-256 fingerprint
9F:8F:99:6D:11:8F:E8:09:CA:B8:3B:D2:F3:09:90:B3:A0:A0:83:E3:A1:9E:E9:FF:13:19:9C:8F:A2:C4:87:AD

Set SSL_CERT_FILE to the verified certificate path before running the optional connector.

Agent entry point

One address for the memory service

Give this address to the agent you already run. The agent can discover the connection details it needs from there. Creating a new connection remains tenant-approved and requires a signed-in Conduir user to issue the one-time setup grant below.

Cortex MaaS base URL
https://18.133.126.110:23443/memory
Your agent asked for a setup grant? That is expected — agents cannot issue their own credentials. Name the agent below and create the connection; you will be shown a one-time grant valid for 15 minutes. Send it to your agent through a secure channel — never paste it into a chat, ticket, or prompt.
Guided setup

Connect the agent you already run

Name the agent and create the connection — only the agent name is required. Recommended settings are applied for everything else and can be adjusted below before you create the one-time setup grant. Your organisation and signed-in identity are applied automatically; there are no identity fields to edit.

Verifying this browser's Conduir session…

2-80 characters; shown in your connection list. This is the only required field.

Recommended settings (applied)
Runs:
Outside Conduir
Memory:
Private to this agent
Access:
Recall and remember
Retention:
Organisation policy
Adjust purpose, boundary, access, or retention

10-500 characters. A recommended purpose is pre-filled — replace it with the agent's real role where you can. Do not paste prompts, credentials, or customer data here.

Where does this agent run?
Memory boundary

Start private unless collaboration is required and a shared memory has already been approved.

Memory access

Grant only what the agent needs for its role.

Retention preference

Your organisation's policy remains authoritative and may shorten the requested period.

Your connections

Connected agents

Every memory connection created for your organisation, newest first. Revoking a connection immediately blocks the agent's memory access and cannot be undone; connecting again requires a new setup grant. Memories the agent saved before revocation remain under your organisation's retention policy and must be deleted separately.

Loading your connected agents…

Working guide · humans and agents

Everything your agent needs to use its connection

Once connected, your agent needs to know the connection steps, the memory tools it can call, and how recall and saving really behave. The full working guide covers all of it — including why the setup grant lasts only 15 minutes while the connection itself is durable until revoked.

Read the working guide
Give your agent the guide directly
/agents/guide.md

The same guide as plain text on this host: your agent can fetch and read it without a browser. No sign-in required; it contains no secrets.

Prepare locally now · governed import remains inactive

Bring your history

Start with a private, browser-local preflight of an earlier agent, chat archive, or project history. It can show bounded record counts, redacted risk indicators, and review suggestions without uploading the source. The governed journey below is the product blueprint; server import, approval, promotion, and recall are not enabled yet.

Governed import target · inactive
  1. 1 · Choose agent

    Select an active connection you are allowed to manage.

  2. 2 · Add source

    Upload a supported export or paste bounded history.

  3. 3 · Choose processing

    Request decisions, summaries, reusable facts, or both.

  4. 4 · Preview

    See exclusions, redactions, citations, and proposed memories.

  5. 5 · Approve

    Select exactly what the agent may recall later.

  6. 6 · Receive receipt

    Keep a record of provenance, retention, and deletion controls.

Local preflight available · import not active

Inspect a history export on this device

No upload

Choose a bounded UTF-8 text, Markdown, JSON, or JSONL export. This advisory check runs in your browser and the source never leaves this device. It shows record counts, redacted risk indicators, and local review suggestions only.

Local checks are incomplete and do not make content safe or trusted. A future active import will scan everything again, require review, and provide lifecycle controls before approved material can become memory. Indicators may miss sensitive material or over-flag harmless text; their counts are not authoritative. Local review suggestions are not approved memories or verified facts. If you download a manifest, it contains redacted excerpts plus explicit no-upload/no-write guarantees, so handle it as sensitive data.
Up to 1,000,000 bytes · at most 2,000 records
Using the connection

A safe memory loop for every agent

Cortex supplies memory, not inference. Your agent remains responsible for its system instruction, model calls, tool decisions, and validation.

1. Recall relevant context before the agent starts its task.
2. Treat recalled material as untrusted context, never as authority.
3. Run the agent with its existing model, tools, prompt, and knowledge sources.
4. After success, remember only a bounded and approved checkpoint.
5. Verify persistence before treating the checkpoint as durable.

Identity is automatic

The signed-in organisation and connection determine the memory boundary. The agent cannot select another owner.

Shared means approved

Entering a shared-memory name never creates access. Existing approval is checked when the connection is created.

The agent keeps its credential

The one-time grant lets the agent generate and retain its own secure credential. Conduir never receives that secret.