CMP42 // PRODUCT // PROCESSESMOVEMENT 03 OF 03
// Movement 03 · Processes · The observable workflow

AUDIT AN AGENT
LIKE A GIT BRANCH.

A process is a named sequence of steps. Some run on humans, some on agents. Each step reads and writes to the graph, each transition is logged, each run is replayable. You see what the agent read, what it wrote, what it cost — and who approved it.

// The problem with invisible automation

AUTOMATION YOU CANNOT INSPECT IS A LIABILITY.

Most CRM-to-AI workflows today are a Zapier hop and a lot of Python. They run, mostly. When the renewal note comes out wrong, nobody can say which input the agent read, which version of the account it saw, or whether a human ever looked at it before it went out. The results are inconsistent, and the inconsistency is invisible.

That is tolerable for a demo. It is not tolerable for client work, for regulated teams, or for anyone who has to explain afterwards why an agent wrote what it wrote.

CMP42 makes agent work reviewable. A process answers four questions for every run: what did it read, what did it write, at what token cost — and would the answer have been different against last month’s snapshot?

// Anatomy of a process

STEPS. HOOKS. RUNS.

Steps

Agents and humans, in sequence

A process is a named sequence of steps. Some run on humans — a partner reviews a case note. Some run on agents — Claude drafts the follow-up. You decide which is which.

Graph

Every step reads and writes the graph

Steps do not pass loose text to one another. They read from and write to the same Context graph, validated against your Models, so each step works from the same source of truth.

Approval

Human-in-the-loop hooks

Put a human approval step wherever judgement matters. The process waits for a person to approve before the next step runs — nothing an agent drafts has to go further unreviewed.

History

Full run history

Every transition is logged. Open any past run and follow it step by step: inputs, outputs, decisions, who or what acted at each point.

Attribution

Every write has an author

Every write records the human or agent that produced it. “The system changed it” is never the answer.

Cost

Token spend per process, per run

Token spend is visible per process and per run, so the cost of an automated step sits next to the work it produced.

Human approvals are the same event on four channels. A step configured as approve notifies the responsible person through whichever channel the workspace has configured: the Podma Workbench with an approve/reject button, the MCP tool process.approve(runId, decision) that an agent can call on someone else’s behalf, a signed approval link emailed to an external manager who is not a CMP42 user, or a chat-client callback (Slack or Teams bot posts the decision into a channel, the button click calls back into CMP42). No embedded widget on this site, no cookies for any of these paths.

// An example process

DRAFTED BY AN AGENT.
APPROVED BY A PARTNER.

A product-led B2B team wants an agent to read a real customer’s history before it drafts the renewal note — not a generic summary that gets the account name wrong. The process on the right describes that workflow.

  1. An agent gathers context from the graph: the deal, the company, the people involved, the latest documents.
  2. The agent drafts the renewal note and writes it to the deal.
  3. A partner reviews it. They can request changes — the agent revises — or approve.
  4. Only the approved note moves on.

The run log shows what that looks like afterwards. Definition format, step names and figures are illustrative; how processes are declared ships with the Alpha and is shaped together with the Design Partners.

processes/renewal-note.yamlIllustrative
process: renewal_note
steps:
  - id: gather
    run:     agent               # e.g. Claude via MCP
    reads:   [deal, company, contact, doc]
  - id: draft
    run:     agent
    writes:  note -> deal
  - id: review
    run:     human               # role: partner
    approve: required            # approval hook
  - id: hand_off
    run:     human
Run history · renewal_noteIllustrative
run renewal_note · deal: "Aspen · Q4 renewal"
  gather    agent:claude    read 7 objects        1,840 tok
  draft     agent:claude    wrote note v1         2,310 tok
  review    human:partner   changes requested
  draft     agent:claude    wrote note v2         1,120 tok
  review    human:partner   approved ✓
  hand_off  human:partner   done
  ─────────────────────────────────────────────────────────
  total                                           5,270 tok
// Replay and review

REPLAY AGAINST LAST MONTH.

Every object in the graph is versioned, so a process run is not a one-off event you can only read about. You can replay any process against a past snapshot and check whether the answer would have been different.

That turns vague worries into questions you can answer: Did the note go wrong because the data was wrong, or because the agent was? If we switch the model or change the step, does last month’s output improve? What exactly did the agent know on the day it wrote this?

If you know how to review a branch, you already know how to review an agent:

In gitIn a CMP42 process
Commit with an authorA write attributed to the human or agent that produced it
DiffWhat a run read and what it changed in the graph
Pull-request reviewA human-in-the-loop approval step
LogThe full run history, transition by transition
Checking out an old commitReplaying a process against a past snapshot
Terminal · replayIllustrative
$ cmp42 replay renewal_note --run <run-id> --snapshot 2026-08-31
  ↺ replaying 6 steps against snapshot 2026-08-31
  Δ draft   note differs from original run  → open diff
// Cost accounting

KNOW WHAT A RUN COSTS.

Agents are not free, and “the AI bill went up” is not a useful finding. Because CMP42 records token spend per process and per run, you can see which workflows are worth automating, which steps are expensive, and what a single approved renewal note actually cost to produce.

CMP42 does not ship its own model — you bring the one you trust. Cost accounting makes its spend visible where the work happens, next to the output and the human who approved it.

Token counts come from two sources. The MCP client itself reports usage on every tool response — that is the primary signal, captured in the run history the moment the tool call completes. In parallel, CMP42 reconciles once a day against the model provider’s usage endpoint (OpenAI, Anthropic, Google) so the numbers survive a client-side drop. Costs are shown two ways: input and output tokens separately, and in euro converted at the pricing snapshot taken at run time — later price changes never distort the historical figures.

// Questions on this

What people ask here

The same answers as in the CRM FAQ, next to the case they belong to.

SAL-01

How do new leads get into CMP42, and what does the agent do?

CMP42 does not collect leads by itself: there is no tracker and no pixel. Leads arrive by import or through the API; in the Alpha, for example, by CSV import for contacts, companies, deals and notes. Lead forms from Meta, Google and LinkedIn are planned to come in labelled with what each of them proves in legal terms.

The agent takes over once a lead has arrived: it checks whether the person is already known, links them to their company and proposes the next step. Every contact carries its legal basis, consent date and consent source (see chapter 3).

Prompts and sources for this question

Decided1
SAL-02

How does an agent qualify leads with CMP42?

The principle is: read first, then ask. Classic qualification flows ask even a three-year customer what their company does. With CMP42 the agent reads the context before the first question: is the person known, which company do they belong to, is there an open deal, what was agreed last time?

A typical flow:

  1. A request comes in, and the address matches an existing contact.
  2. Over MCP, the agent reads the company, the open deal and the note from the last call.
  3. It skips the questions that have already been answered and asks only the open one.
  4. The answer is written back as a qualification record, linked to contact, company and deal.
  5. A Process puts it in front of the deal owner for approval.

Note that CMP42 is the layer the agent reads, not the chatbot itself. There is no widget and no messaging integration; your channel stays yours. Structured assessments, BANT for people and MEDDICC for deals, are planned for the next version.

Prompts and sources for this question

Decided1
SAL-06

How is a forecast produced?

A forecast predicts revenue from open and expected deals. With CMP42 you ask your agent, for example: “Which deals are likely to close in Q4, and which are at risk?” The agent reads deal values, stages, recent activities and notes, and justifies its estimate with the records it read.

The judgement stays with people. A Process can put the forecast up for approval, and the decision is recorded with its author. If the Model or the flow changes, the forecast can be rerun against an earlier state of the data and compared.

Prompts and sources for this question

Decided1
MKT-03

How does lead nurturing work with an agent?

Nurturing means keeping a relationship alive with relevant content until the contact is ready to buy. The weak point of generic AI text is that it does not know the history. With CMP42 the agent reads the account before every draft: who owns the deal, which documents exist, what the latest notes say.

The effect grows over time. Every approved note becomes part of the graph that the next run reads, so the second message starts with more knowledge than the first.

Prompts and sources for this question

MKT-07

How does CMP42 make sure we contact people lawfully?

CMP42 asks two questions in a fixed order. First: may anyone contact this person at all, on which channel, in which country and for which purpose? Second: is the contact worth making? That way a campaign does not fail at the send button after four weeks of work.

Already in the Alpha (since October 2026) Planned for the next version
Legal basis, consent date and consent source on every contact Reachability check per channel, country and purpose, with a reason
No recipient list without an eligibility check, 1 million records measured in 2.1 s Live count of eligible recipients while you define an audience
Eligibility check at any time on any view Jurisdiction templates, starting with the EU/EEA, Switzerland and the UK

CMP42 does not decide for you whether a campaign is lawful. You remain the controller; the templates support your documentation under GDPR Art. 5(2) and Art. 7(1) and are not legal advice.

Prompts and sources for this question

Decided1
MKT-01

Can CMP42 replace marketing automation?

Not entirely, and that is deliberate. Sequences, sending and opt-out management are not part of the Alpha. CMP42 takes the part that comes before them, the part classic tools often reach too late: it checks who may be contacted at all and why, and it rejects a recipient list that breaks the rules.

The work is split like this: the agent proposes segments, content and the next step. CMP42 supplies the context and checks the rules. A person approves. Your existing tool still does the sending.

Prompts and sources for this question

Decided and on the way. Say which one you need first:

  • Open - you decide1Source records with provenance
  • Open - you decide1Approved forecast snapshots
  • Open - you decide1Follow-up sequences
  • Open - you decide1Event-driven processes
  • Open - you decide1Record history over MCP
  • Open - you decide1Reminders before retention ends

All 43 questions in the CRM FAQ

¹ Planned for a future release.