Platform

Specialized agents that own work end to end.

Not a task when asked. The whole engagement, with your experts in the loop. Build agents from your specialists' expertise, prove them on your firm's past work, and improve them from every expert correction.

Talk to us

sell-side

Welcome back, Roland

Conversations

116

Costs

$189.67

Deployments

Needs attentionView all

Data room sync misses re-uploaded files New

The agent read the first data room sync and missed 14 files the seller re-uploaded, citing superseded financials.

Agent struggle18 conversationsLast seen 24 min ago

Financial summary mixes fiscal and calendar years

The agent put FY and calendar-year revenue in one table, then corrected it only after the associate flagged the gap.

Agent struggle9 conversationsLast seen 3 hr ago

Datasite searches omit the project ID

The agent searched Datasite without a project ID, causing validation errors before recovering it from the deal folder.

Agent struggle6 conversationsLast seen 1 day ago

Runtimes

cim-agent

42 conversations$18.422% tool error rate

The work, done to your firm's standard.

Each agent works in your firm's templates and house style, and cites every figure and finding to its source, so your team reviews the work instead of redoing it.

Atlas CIM, key figuresDraft

M&A advisoryFor deal team review

Figure	Value	Source
FY25 revenue	$186m, +24%	DR 3.1, p. 12
Net revenue retention	118%	DR 3.4, p. 6
Top ten customers	19% of revenue	DR 3.2, p. 4
Gross margin bridge	Awaiting data	Management

Covenant comparison4 deviations

Private creditLedgerline vs your terms

Term	Standard	Ledgerline	Source
Max leverage	5.50x	6.25x	§7.11
RP basket	$25m	$40m	§7.06
Add-back cap	20%	Uncapped	§1.01
Equity cures	2 in 4Q	3 in 4Q	§8.04

Diligence grid2 flagged

Law214 contracts reviewed

Contract	Change of control	Source
Northlake MSA	Consent required	p. 14
Supply agreement	None	p. 3
Reseller agreement	Termination right	p. 9
Office lease	Notice only	p. 22

Recipes

Your firm's methods, as agents.

Each agent is a recipe: your firm's method written down, with the skills and tools it uses and the standards it is judged by. Your engineers can read it, change it and own it.

Learn about Recipes

From data room to first-draft CIM: read every document, build the financial summary and the equity story in your house style, and cite each claim to the page for the deal team's review.

Covenants, baskets and EBITDA definitions compared across every agreement in the book, each term cited to its section and every deviation from your standard terms flagged for the deal team.

Clause by clause across the data room: change of control, assignment, exclusivity and termination in one grid, every finding cited to the page, then the first draft of the diligence report.

Data Room to CIM

From data room to first-draft CIM: read every document, build the financial summary and the equity story in your house style, and cite each claim to the page for the deal team's review.

.introspection

cim-agent.yaml

.pi

mcp.local.example.json

agents

agent.yamlM

data-room-reader.yaml

financials.yaml

cim-drafter.yaml

skills

house-style

mnpi-handling

judges

citation_accuracy.yaml

house_style.yaml

evals

carve-out-sale.yaml

incomplete-data-room.yaml

SYSTEM.md

package.json

1name: agent
2description: Sell-side co-worker, data room to first-draft CIM.
3ai:
4  model: anthropic/claude-opus-4-8
5  thinking_level: high
6tools:
7  - read
8  - bash
9  - write
10mcp:
11  datasite:
12    include: ["*"]
13  sharepoint:
14    include: ["*"]
15  capiq:
16    include: ["*"]
17subagents:
18  - data-room-reader
19  - financials
20  - cim-drafter
21skills:
22  - house-style
23  - mnpi-handling
24system_instructions:
25  mode: append
26  content: |
27    # Role: Sell-side associate
28    Read the whole data room, build the financial
29    summary, and draft the CIM in the house style.
30    Cite every figure to its page; leave a gap
31    marked for the deal team rather than guess.

Credit Agreement Review

Covenants, baskets and EBITDA definitions compared across every agreement in the book, each term cited to its section and every deviation from your standard terms flagged for the deal team.

.introspection

credit-agent.yaml

.pi

mcp.local.example.json

agents

agent.yamlM

agreement-parser.yaml

covenant-extractor.yaml

comparison-grid.yaml

skills

leveraged-finance-terms

credit-memo-format

judges

definition_fidelity.yaml

section_citation.yaml

evals

amended-and-restated.yaml

covenant-lite.yaml

SYSTEM.md

package.json

1name: agent
2description: Credit co-worker, agreements in, comparison grids out.
3ai:
4  model: openai/gpt-5.5
5  thinking_level: high
6tools:
7  - read
8  - bash
9  - write
10mcp:
11  imanage:
12    include: ["*"]
13  sharepoint:
14    include: ["*"]
15subagents:
16  - agreement-parser
17  - covenant-extractor
18  - comparison-grid
19skills:
20  - leveraged-finance-terms
21  - credit-memo-format
22system_instructions:
23  mode: append
24  content: |
25    # Role: Credit analyst
26    Extract covenants, baskets and EBITDA
27    definitions from each agreement, compare them
28    to the fund's standard terms, and cite every
29    term to its section. Flag, never interpret away.

Due Diligence Review

Clause by clause across the data room: change of control, assignment, exclusivity and termination in one grid, every finding cited to the page, then the first draft of the diligence report.

.introspection

diligence-agent.yaml

.pi

mcp.local.example.json

agents

agent.yamlM

document-triage.yaml

clause-review.yaml

report-drafter.yaml

skills

firm-playbook

privilege-handling

judges

clause_recall.yaml

citation_accuracy.yaml

evals

buried-anti-assignment.yaml

scanned-contracts.yaml

SYSTEM.md

package.json

1name: agent
2description: Diligence co-worker, data room to diligence report.
3ai:
4  model: anthropic/claude-opus-4-8
5  thinking_level: high
6tools:
7  - read
8  - bash
9  - write
10mcp:
11  imanage:
12    include: ["*"]
13  intralinks:
14    include: ["*"]
15subagents:
16  - document-triage
17  - clause-review
18  - report-drafter
19skills:
20  - firm-playbook
21  - privilege-handling
22system_instructions:
23  mode: append
24  content: |
25    # Role: Diligence associate
26    Review every contract against the firm's
27    playbook, record each finding in the grid with
28    its page, and draft the report for the partner.
29    Nothing reaches the client without their read.

Runtime

Built for work that takes weeks.

Runtime carries each engagement from first signal to final draft, for as many weeks as it takes. It reacts as filings, replies and date changes arrive, schedules the next step, waits on what others owe, and records every action so a partner can replay what it did and why. It runs under its own identity with only the access the work needs.

Its own identity

Every agent acts under an identity scoped to its work, never a person's login.

Your systems, connected

Data rooms, document systems and market data, with only the access each agent needs.

Weeks on one engagement

Work carries on through interruptions and restarts without losing its place.

Remembers how you work

House style, past corrections and each engagement's history, from one session to the next.

Operator

Your firm's automated AI researcher.

Operator studies where your agents fall short, learns from every expert correction, and tests each fix against your firm's standards before it ships. Rented products can't run this loop for you, because the loop belongs to the vendor. On the Platform, the agents own the work and the firm owns the loop.

Live agent

Fix A

Fix B▲ +9.2pt

How we work.

We turn your specialists' expertise into production agents. Start with a use case we have built before, prove it on your own past deals and matters, then expand across your teams and clients.

Embedded

We work alongside your specialists and engineers to define success, connect your systems, and deliver a production outcome.

Platform

Your engineers build and operate recipes on Introspection, with Operator helping them improve agents from production evidence.

Private agent cloud, private intelligence.

Your data stays in your cloud and never trains anyone else's model. Every agent works under your firm's identity, records and approvals.

Security and deployment

Your cloud or on-prem
Deploy the whole stack in your own cloud or on your own servers, with keys you hold. Your data never leaves.
No passwords in the agent
Each agent gets only the access its work needs. Credentials are swapped in at the gateway, so the agent never sees one.
Every action on the record
Conversations, tool calls, costs and approvals, by agent, team and client.
Approval before it leaves
Nothing client-facing goes out without a person's yes.

Own your institutional intelligence.

Talk to us

Specialized agents that own work end to end.

Welcome back, Roland

The work, done to your firm's standard.

Your firm's methods, as agents.

Built for work that takes weeks.

Your firm's automated AI researcher.

How we work.

Embedded

Platform

Private agent cloud, private intelligence.

Your cloud or on-prem

No passwords in the agent

Every action on the record

Approval before it leaves

Own your institutional intelligence.