This is a demo with sample data · Book a demo with your data →

TALLYGRAPHAI AGENT SECURITYCreate a free account

Live demo

What a team sees about its coding agents: which secrets they exposed, which sessions were risky and what they cost. The analysis runs on each developer machine and only uploads signals, never transcripts.

Agent sessions

Sample sessions. TallyGraph monitors and flags: it does not block agent actions.

Session summary and associated spend (retention window: 30 days)

Local session risk and gateway call risk are reported separately and never added together, since they can describe the same usage.

Spend on risky sessions

$12.45

3 sessions with findings · 0 unpriced · 0 without matched usage

Observed trifecta rate

20.0% (1/5 with full coverage · 0 incomplete)

Sessions with trifecta / sessions with full trifecta coverage

Exposures / MCP drift

3 / 1

Local sensitive patterns / MCP definition changes

Suspected injection attempts

1

Deterministic heuristic (5 tool call window)

Live shadow decisions

1 req / 1 eval

Pre-execution (separate from retrospective replay)

Retrospective / Enforce decisions

Retro: 0/0 · Enf: 0/0

Transcript replay vs runtime gate

No findings does not mean guaranteed safety and does not rule out paths outside coverage. Findings indicate observed risk conditions, not confirmed compromise or misconduct.

Policy status and installation acknowledgement (v1)

Requested mode: shadow · Effective mode: shadow · Adapters with verified interception: none (monitoring only / unavailable for enforce)

Policy acknowledgement per enrolled installation
InstallationAgentAcknowledged modeAcknowledged versionEffective status
inst_demoAna7Xv2Qp8Ls4claudeno acknowledgement—Monitoring only
inst_demoLuis4Nm6KrT9wcodexno acknowledgement—Monitoring only
inst_demoMarta8Ls4Qp2claudeno acknowledgement—Monitoring only
inst_demoLuis2Hc8Zr5yclaudeno acknowledgement—Monitoring only
inst_demoAna3Jt6Wd1Fqcodexno acknowledgement—Monitoring only

Named observed sessions (5)

List of observed sessions with developer identity, spend, coverage and sanitized findings
DeveloperAgent / RefObserved windowSpend and price coverageTrifectaExposure / Drift / InjectionMonitoring mode
Ana
inst_demoAna7Xv2Qp8Ls4
claude
ses_demo1a2b3c4d5e6f7a8b9c0d
Start: 2026-11-20T04:00:00.000Z
End: 2026-11-20T04:48:00.000Z
Expires: 2026-12-20T04:00:00.000Z
$7.42 (full coverage)
High severity
Detector coverage: complete
Exp: 2 · Drift: 0 · Inj: 0
Timeline (3) and versions
Detector coverage and versions: exposure=v1(complete), trifecta=v1(complete), mcp_drift=v1(complete), injection=v1(complete), shadow_policy=v1(complete), runtime_gate=v1(unavailable) · Policies: v1
  1. 2026-11-20T04:09:00.000Z · Exposure (credential, rule credential_pattern): 2 (High severity)
  2. 2026-11-20T04:21:00.000Z · Trifecta (high): private data + untrusted input + outbound capability (High severity)
  3. 2026-11-20T04:22:00.000Z · Policy v1 (shadow): approval_required on network (#42) (Medium severity)
Live shadow (does not block)
Luis
inst_demoLuis4Nm6KrT9w
codex
ses_demo2b3c4d5e6f7a8b9c0d1e
Start: 2026-11-19T22:00:00.000Z
End: 2026-11-19T22:35:00.000Z
Expires: 2026-12-19T22:00:00.000Z
$3.15 (full coverage)
No trifecta finding
Detector coverage: complete
Exp: 0 · Drift: 1 · Inj: 1
Timeline (2) and versions
Detector coverage and versions: exposure=v1(complete), trifecta=v1(complete), mcp_drift=v1(complete), injection=v1(complete), shadow_policy=v1(complete), runtime_gate=v1(unavailable) · Policies: v1
  1. 2026-11-19T22:04:00.000Z · MCP drift (description_changed): v1 -> v2 (Medium severity)
  2. 2026-11-19T22:16:00.000Z · Suspected injection (high): tool_output (#17) -> new_destination (#19) (High severity)
Monitoring only (shadow)
Marta
inst_demoMarta8Ls4Qp2
claude
ses_demo3c4d5e6f7a8b9c0d1e2f
Start: 2026-11-19T07:00:00.000Z
End: 2026-11-19T07:22:00.000Z
Expires: 2026-12-19T07:00:00.000Z
$1.88 (full coverage)
No trifecta finding
Detector coverage: complete
Exp: 1 · Drift: 0 · Inj: 0
Timeline (1) and versions
Detector coverage and versions: exposure=v1(complete), trifecta=v1(complete), mcp_drift=v1(complete), injection=v1(complete), shadow_policy=v1(complete), runtime_gate=v1(unavailable) · Policies: v1
  1. 2026-11-19T07:06:00.000Z · Exposure (credential, rule credential_pattern): 1 (High severity)
Monitoring only (shadow)
Luis
inst_demoLuis2Hc8Zr5y
claude
ses_demo4d5e6f7a8b9c0d1e2f3a
Start: 2026-11-19T03:00:00.000Z
End: 2026-11-19T04:04:00.000Z
Expires: 2026-12-19T03:00:00.000Z
$9.60 (full coverage)
No trifecta finding
Detector coverage: complete
Exp: 0 · Drift: 0 · Inj: 0
Timeline (0) and versions
Detector coverage and versions: exposure=v1(complete), trifecta=v1(complete), mcp_drift=v1(complete), injection=v1(complete), shadow_policy=v1(complete), runtime_gate=v1(unavailable) · Policies: v1

No signal events in this review.

Monitoring only (shadow)
Ana
inst_demoAna3Jt6Wd1Fq
codex
ses_demo5e6f7a8b9c0d1e2f3a4b
Start: 2026-11-18T08:00:00.000Z
End: 2026-11-18T08:18:00.000Z
Expires: 2026-12-18T08:00:00.000Z
$0.94 (full coverage)
No trifecta finding
Detector coverage: complete
Exp: 0 · Drift: 0 · Inj: 0
Timeline (0) and versions
Detector coverage and versions: exposure=v1(complete), trifecta=v1(complete), mcp_drift=v1(complete), injection=v1(complete), shadow_policy=v1(complete), runtime_gate=v1(unavailable) · Policies: v1

No signal events in this review.

Monitoring only (shadow)

What they cost

Spend for the same team across agents and AI APIs. Refreshes every hour.

Spend by service

This monthView records

Monthly trend

6 monthsView records
Closed periodsCurrent month (partial)

Month forecast

View records

Forecast based on activity so far this month.

~$217.91

$195.27 spent · +24.0 % vs previous month

Recommendations

7 findingsView decisions
Agent · high confidence~$144.55/mo

Consider MiMo V2.5 Pro API

The agent repriced the same tokens from the last 30 days at official prices. MiMo V2.5 Pro API could lower the estimated monthly cost from $149.53 to $4.97, with 78.9% on SWE-bench Verified as reported by the provider. Check quality and compatibility before migrating.

inspect_tokens → price_token_workload → compare_benchmarks · human approval required

View official pricingView benchmark
  • Idle subscription~$30.00/mo

    MiniMax

    MiniMax recorded no usage in the last 30 days, though it has 8 historical records. Its monthly cost is $30.00.

  • Duplicate service~$20.00/mo

    Anthropic (Console)

    There are 2 overlapping active Anthropic subscriptions; Anthropic (Console) is the cheapest, at $20.00 a month.

  • Uncapped usage~$4.14/mo

    OpenAI

    OpenAI usage went from $0.26 last month to $4.41 this month (16.63x); the increase is $4.14.

View all in Decisions

Services

This month
% of total
Anthropic$139.64
71.5 %
MiniMax$30.00
15.4 %
Anthropic (Console)$20.00
10.2 %
OpenAI$4.41
2.3 %
Moonshot (Kimi)$0.55
0.3 %
DeepSeek$0.41
0.2 %
Z.ai (GLM)$0.21
0.1 %
Google AI$0.05
0.0 %
  • Anthropic

    Claude Max 5x

    $100.00

    in 17 days

  • Anthropic (Console)

    Claude Pro

    $20.00

    in 24 days