Governance for AI agents

Every agent action, on the record.

What your AI agents read, run and change, starting with Claude Code and Codex across every developer, and who wrote each line.

Book a Demo See how it works ↓

Encrypted in transit and at rest
Known secrets never stored
Zero-effort install

sightline.yourcompany.com

Active developers24+2 vs previous 7 days

Sessions1,284+118 vs previous 7 days

Tool calls38.6K+4.2K vs previous 7 days

Files edited912+86 vs previous 7 days

Tests run3,42099% passed+310 vs previous 7 days

What agents did, per day

ReadsEditsCommandsFetchesAgents & toolsShow table

Oct 6

Edits1,670

Commands1,830

Fetches530

Agents & tools380

Total7,600

Needs attention

Denied tool calls17needs a look

Failed test runs23needs a look

Upstream errors2needs a look

Risky commands4needs a look

New MCP servers2needs a look

Unidentified sessions1needs a look

Flagged results3needs a look

MCP servers

Server	Developers	First seen	Last seen	Calls	Status
github	dana@yourcompany.comsam@yourcompany.compriya@yourcompany.com	2026-08-04 09:12	2026-10-08 13:52	1,840
linear	lee@yourcompany.comjordan@yourcompany.com	2026-08-19 11:40	2026-10-08 13:02	612
notion-export	jordan@yourcompany.com	2026-10-06 16:25	2026-10-08 10:47	38	new
sentry	priya@yourcompany.comalex@yourcompany.com	2026-10-08 13:24	2026-10-08 13:51	21	new

ActivityFeed

14:01:37lee@yourcompany.comClaude Codeclaude-sonnet-4-5Session 2b81f0c4…91aaTurn 622,610 in, 1,442 out, 6.2s

support-toolsticket-search

Paginate the ticket search endpoint

1 file changed

▸ api/tickets/search.ts

3 reads, 1 command

13:59:12jordan@yourcompany.comCodexgpt-5-codexSession e40d77a1…3c08Turn 29,870 in, 812 out, 4.9s

data-pipelinemain

Export the weekly churn table to the notebook

2 reads, 1 mcp call

ActivityFeedSession 7c41e2a9…4d2e

Session 7c41e2a9-58b0-4c1e-9f3a-b6d2c0e84d2eClaude CodeClaude Code session 0199c4f2-1b7e-7a30Repo billingBranch fix-prorationHead 5d1e07b3

Commits recorded during this sessionFix proration bug in invoice totals 8f3c2a91

readsearcheditcommandtestfetchskillsubagentmcptool

prompt14:02:11 Fix proration bug in invoice totals
readsrc/billing/invoice.tsok
testpnpm test --filter billingfailed
reveal output (2,184 bytes)
edit▸ src/billing/invoice.tsEdit+15 −3history
testpnpm test --filter billingpassed
reveal output (1,032 bytes)
edit▸ .env.productionWriteblocked by Protected paths
14:02:58claude-sonnet-4-5in 18,240out 2,1068.4sreveal system prompt

GovernAlerts

Policy gates24 machines on the current policy.

Rule	Developer	Repo	Severity	Detail	Count	Last seen
Protected paths	dana Claude Code	billing	critical	.env.production	8	1m ago	Resolve
Deny command	sam Codex	infra	critical	git push --force	6	18m ago	Resolve
Secrets sent to the model	priya Claude Code	web-app	critical	AWS access key	3	2h ago	Resolve
Personal data sent to the model	lee Claude Code	support-tools	serious	customer email address	5	4h ago	Resolve
Approved MCP servers only	jordan Codex	data-pipeline	warning	mcp: notion-export	2	7h ago	Resolve
Deny command	alex Claude Code	infra	critical	terraform destroy	1	1d ago	Resolve

ProveAttribution

How to read this pageShow

AI sharei31%57,040 of 184,000 lines seen; another 15% AI, then edited

Written by AIi31%57,040 lines

AI, then editedi15%27,600 lines

Written by peoplei48%88,320 lines

Carried overi6%11,040 lines

Commits recordedi2,140first on 2026-09-08

AI share by developer

dana@yourcompany.com

42%

priya@yourcompany.com

37%

sam@yourcompany.com

33%

lee@yourcompany.com

28%

jordan@yourcompany.com

24%

alex@yourcompany.com

19%

AI share by repository

web-app

39%

billing

32%

support-tools

29%

infra

22%

data-pipeline

20%

Commits

AIAI then editedPeopleCarried overNot seen

Commit	Repository	Author	When	Lines added	Who wrote iti	Evidencei
Fix proration bug in invoice totals 8f3c2a91 Claude Code	billing fix-proration	dana	1m ago	18	AI 61% · AI then edited 22% · People 17%	Completei
Add audit fields to the refunds table 41d07be5 Codex	billing refund-audit	priya	1h ago	96		Completei
Paginate the ticket search endpoint c92e5f10 Claude Code	support-tools main	lee	3h ago	158		Completei
Tighten CSP headers on checkout 0be6d4a7 Claude Code	web-app csp	sam	5h ago	41		Completei

Agents write a real share of your code. Can you answer these?

Did an agent read your API keys?
Secrets are flagged before they reach the model, and protected files are blocked on the developer’s machine.
How much of your code did AI write?
Line-level attribution for every commit.
Which files did agents touch?
Every read, edit and command, by developer.
What were they told to do?
The prompt behind every session.

From prompt to commit, on the record.

Observe

See every action, by every agent.

A live feed of every turn: what each agent read, ran and changed, with session replay and per-repo and per-file history. Denied calls, failed tests, risky commands, new MCP servers and flagged results surface on their own.

Explore Observe →

Repository

billing

Today

D
dana · Claude Code
Fix proration bug in invoice totals
- src/billing/invoice.ts +15 −3
14:02
priya connected a new MCP server: sentry New
13:24
P
priya · Codex
Add audit fields to the refunds table
- db/migrations/0042_refund_audit.sql +38 −0
- src/billing/refunds.ts +58 −6
12:58
S
sam · Claude Code
Retry declined card payments
- src/billing/dunning.ts +22 −4
12:31

Govern

Stop what shouldn’t happen, before it does.

Rules that alert or block: deny a command, protect a path, catch secrets or personal data before they reach the model, allow only approved MCP servers. Blocks happen on the developer’s machine, in Claude Code and Codex.

Rules reference →

Blocked: write to .env.production

dana · billing · Claude Code

Blocked

Write .env.productionStopped before the file was written

Rule: Protected paths
Time: 14:02
Seen: 8 times

Rules

Protected pathsOn
Secrets sent to the modelOn

24 machines on the current policy.

Prove

Prove who wrote what, line by line.

For every commit, the share written by AI, by AI then edited, by people, or carried over, by developer, repo and model. Export it to CSV for the people who ask.

Explore Prove →

Fix proration bug in invoice totals

8f3c2a91 · billing · fix-proration

AI
Claude Code · 11 lines
61%
D
AI then edited
dana reworked 4 lines
22%
D
People
dana · 3 lines
17%

Indicators

Catch the instructions hiding in what your agents read.

Every result a tool hands back to an agent is checked for text that tries to instruct it, from a fetched page to a command’s output. SightLine flags it by category and keeps only the category, never the text.

See what’s recorded →

Indicators

Last 7 days · every tool result

Flagged results: 71tool results
Share of all results: 1 in 1027,250 results
Most common: tool coercion67 results

By categoryShare of 71 flagged

instruction override4
Phrases like “ignore all previous instructions”, “you are now a…” or “do not tell the user”.
role injection2
A line beginning “SYSTEM:” or “Assistant:”, chat-template markers, or a <system> tag inside content.
hidden text1
Five or more invisible zero-width characters, a right-to-left override, or an HTML comment addressed to the assistant.
tool coercion67
A “curl … | sh” pipeline, a request to send a key, token or password, or a read of ~/.ssh, .env or cloud credentials.
encoded payload0
A run of 200 or more base64 characters in a fetched page, a search result or an MCP reply.

Everything your agents do passes through one line.

SightLine sits between your developers’ agents and your record. Activity goes in, risky actions stop at the line, and a clean, attributed record comes out the other side.

Every agent, every laptop

Claude Code and Codex sessions across your team. Omega sets up capture once, for the whole team.

The line

Recorded and checked

Each action is recorded, and your rules block what shouldn’t happen, right on the developer’s machine.

Out

One clean record

The feed, alerts and line-by-line attribution, from the first session.

Built for the people who ask.

An audit trail your auditors can use.

Every agent action is recorded and every view of someone else’s stored content is logged. Rule changes are kept with who changed what, and when.

Rules & compliance →

Access log

Every view of someone else’s stored content

A
admin opened prompt
dana · 7c41e2a9…4d2e
14:06Oct 8
A
admin opened edit
dana · b3e90c17…5a21
14:06Oct 8
S
security opened tool input
priya · 91c3a5de…07b2
11:42Oct 8

Know what AI is doing in your engineering org.

Adoption, sessions and the AI share of every commit, by developer, repo and model.

See the product →

AI share

Last 30 days · every repository

31%57,040 of 184,000 lines seen

web-app39%
billing32%
support-tools29%
infra22%
data-pipeline20%

See the code, files and commands agents touch.

A live feed, session replay and per-file history, plus rules that block on the developer’s machine.

Rules reference →

src/billing/invoice.ts

Claude Code edit · 8f3c2a91

+15 −3

…

AIconst { start, end } = inv.period;

AI// inclusive of the last day

AIconst days = end.diff(start, "day") + 1;

danaif (days <= 0) return 0;

Editedconst total = inv.daysIn(start);

Editedreturn round(inv.amount * days / total);

…

AI · Claude Code
AI then edited · dana
People · dana

Built for the org.

Team setup
Push one setup to every developer.
- Shared skills
- Slash commands
- CLAUDE.md baseline
v13 · published
1. v13Published todayCurrentAdded the /release-notes command
2. v122 weeks agoUpdated the CLAUDE.md baseline
3. v11Last monthAdded a shared code review skill
Your company’s SSO
Sign in with your company’s single sign-on. No separate passwords.
Org-wide rules
Set a rule once. It applies to every developer.
- Protected pathsOn
- Secrets sent to the modelOn
Audit trail
Every view of someone else’s stored content is logged.
- admin@ opened dana’s prompt · 14:06
- Rule changed: Protected paths · 13:51
Adoption and rollback
See who’s on the latest setup. Roll back in one click.
24 of 24 machines currentRoll back to v12
Claude Code + Codex
One record across both agents.
Claude CodeCodex

Deployment options
	Hosted by OmegaFor growing teams	In your cloudFor large and regulated organizations
Single-tenant	Yes	Yes
Behind your SSO	Yes	Yes
Where records live	An instance Omega runs	Your infrastructure, in any cloud environment you run
Who operates it	Omega	Omega, in your account

Learn more →

We keep receipts. Not your secrets.

What SightLine records

Prompts developers send to their agents
The text of agent edits
Commands and tool calls
A short excerpt of each tool result
Token counts
File names and commit details (author, message subject, who wrote each line)

Read the security overview →

Your auditor’s favorite tab.

Evidence for your SOC 2 and ISO 27001 audits: SightLine records what auditors ask about when AI agents touch production code, from who changed what to who saw what and which controls were in force.

SOC 2 CC8 · ISO 27001
Change management
Which agent changed which files, line-level authorship on every commit, and each rule change with before and after values.
SOC 2 CC6 · ISO 27001
Logical access
Admin and developer roles, sign-in through your SSO, and every view of someone else’s stored content with who, whose and when.
SOC 2 CC7 · ISO 27001
System monitoring
Blocked actions, alerts with first and last seen and resolution, and secret and personal-data detection.
Policy
Policy in force
Whether each machine’s rules are current, the rules version, and the team setup’s version history.

SightLine isn’t itself SOC 2 or ISO 27001 certified. It produces evidence for your audit, exportable as CSV or JSON.

See the full evidence map →

Not just for engineering.

Any AI agent your company runs, from support bots to research agents to internal tools, can run through SightLine and get the same record of every action, the same rules and the same audit trail.

Support bots
Research agents
Internal tools
Coding agents

Talk to us about your agents →

Built by Omega.

Omega Business Solutions is an AI implementation firm in Knoxville, Tennessee. We built SightLine because we put AI agents to work for our clients and needed to account for every action they take.

Visit Omega →

See your agents clearly.

A 30-minute demo. We’ll show SightLine running on a real codebase.

Book a Demo

Every agent action, on the record.

Agents write a real share of your code. Can you answer these?

Did an agent read your API keys?

How much of your code did AI write?

Which files did agents touch?

What were they told to do?

From prompt to commit, on the record.

See every action, by every agent.

Stop what shouldn’t happen, before it does.

Prove who wrote what, line by line.

Catch the instructions hiding in what your agents read.

Everything your agents do passes through one line.

Every agent, every laptop

Recorded and checked

One clean record

Built for the people who ask.

An audit trail your auditors can use.

Know what AI is doing in your engineering org.

See the code, files and commands agents touch.

Built for the org.

Team setup

Your company’s SSO

Org-wide rules

Audit trail

Adoption and rollback

Claude Code + Codex

We keep receipts. Not your secrets.

What SightLine records

Your auditor’s favorite tab.

Change management

Logical access

System monitoring

Policy in force

Not just for engineering.

Built by Omega.

See your agents clearly.