Skip to content

Pivot positioning: observability → cost control #243

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
bmdhodl merged 1 commit into from
Mar 8, 2026
Merged

Pivot positioning: observability → cost control #243

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 9 additions & 5 deletions ops/00-NORTHSTAR.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,19 +1,23 @@
# AgentGuard SDK — North Star
# AgentGuard — North Star

## What it is

A zero-dependency Python SDK that gives AI agent builders runtime guards — hard budget limits, loop detection, and timeout enforcement that kill agents mid-run before they cause damage.
An open-source Python SDK + hosted dashboard that gives AI teams cost control over their agents — hard budget limits, loop detection, spend visibility, and policy management that stop runaway AI bills before they happen.

**SDK:** Zero-dependency runtime guards that kill agents mid-run when they exceed spend limits.
**Dashboard:** The hosted control surface — see spend, set limits, get alerts when things go wrong.

## Who it's for

Python developers building autonomous agent systems with LangChain, LangGraph, CrewAI, or raw OpenAI/Anthropic APIs — especially those moving agents from prototype to production.
Small AI teams (1-10 engineers) building autonomous agent systems with LangChain, LangGraph, CrewAI, or raw OpenAI/Anthropic APIs — especially those moving agents from prototype to production and needing predictable AI costs.

## What problem it solves

Agents fail silently. They loop, overspend, and hang. AgentGuard intercepts these failures at runtime with structured tracing and guards that raise exceptions — not dashboards that show you the damage after it happened.
Agents fail silently. They loop, overspend, and hang. AgentGuard intercepts these failures at runtime with hard budget caps, loop detection, and timeout enforcement. The hosted dashboard gives teams spend visibility and policy management without building infrastructure.

## Non-goals

1. **Not a framework.** We don't orchestrate agents. We guard whatever framework you already use.
2. **Not an observability platform.** We emit structured traces (JSONL) but don't store, query, or visualize them. Sinks push traces elsewhere (files, HTTP, OpenTelemetry).
2. **Not a full observability platform.** We are not LangSmith, Langfuse, or Helicone. We don't do prompt management, eval suites, or deep trace exploration. We control costs and stop runaways.
3. **Not a prompt engineering tool.** We don't evaluate prompt quality or optimize outputs. We stop agents from burning money and looping.
4. **Not enterprise governance.** No RBAC, audit logs, or compliance features in V1. We serve small teams first.
120 changes: 39 additions & 81 deletions site/compare.html
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,25 +3,25 @@
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>AgentGuard vs LangSmith vs Langfuse vs Portkey — AI Agent Cost Tracking Comparison</title>
<meta name="description" content="Compare AgentGuard with LangSmith, Langfuse, and Portkey. Runtime budget enforcement, loop detection, and cost tracking for AI agents. Free, zero dependencies, MIT licensed." />
<title>Why AgentGuard for AI Cost Control? — vs LangSmith, Langfuse, Portkey</title>
<meta name="description" content="Other tools show you the damage after it happens. AgentGuard stops it. Compare budget enforcement, cost tracking, and spend visibility side by side." />
<link rel="canonical" href="https://agentguard47.com/compare.html" />
<link rel="icon" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 100 100'><text y='.9em' font-size='90'>🛡️</text></svg>" />

<!-- Open Graph -->
<meta property="og:type" content="website" />
<meta property="og:url" content="https://agentguard47.com/compare.html" />
<meta property="og:title" content="AgentGuard vs LangSmith vs Langfuse vs Portkey" />
<meta property="og:description" content="The only tool that kills agents mid-run when they exceed spend limits. Compare features, pricing, and integrations." />
<meta property="og:title" content="Why AgentGuard for AI Cost Control?" />
<meta property="og:description" content="Other tools show you the damage after it happens. AgentGuard stops it. Compare budget enforcement and spend visibility." />
<meta property="og:image" content="https://opengraph.githubassets.com/1/bmdhodl/agent47" />

<!-- Twitter Card -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="AgentGuard vs LangSmith vs Langfuse vs Portkey" />
<meta name="twitter:description" content="The only tool that kills agents mid-run when they exceed spend limits. Compare features, pricing, and integrations." />
<meta name="twitter:title" content="Why AgentGuard for AI Cost Control?" />
<meta name="twitter:description" content="Other tools show you the damage after it happens. AgentGuard stops it. Compare budget enforcement and spend visibility." />

<!-- SEO keywords -->
<meta name="keywords" content="langsmith alternative, langfuse alternative, ai agent cost tracking, agent budget enforcement, ai agent observability, langchain cost tracking" />
<meta name="keywords" content="langsmith alternative, langfuse alternative, ai agent cost control, agent budget enforcement, ai agent spend tracking, langchain cost tracking, ai agent budget limits" />

<script defer src="/_vercel/insights/script.js"></script>
<style>
Expand Down Expand Up @@ -118,8 +118,8 @@
<span>Compare</span>
</nav>

<h1>AgentGuard vs LangSmith vs Langfuse vs Portkey</h1>
<p class="muted">The only AI agent observability tool with runtime budget enforcement. Compare features, pricing, and integrations side-by-side.</p>
<h1>Why AgentGuard for AI cost control?</h1>
<p class="muted">Other tools show you the damage after it happens. AgentGuard stops it. Compare budget enforcement, cost tracking, and spend visibility side by side.</p>

<div class="cta">
<a class="btn" href="https://github.com/bmdhodl/agent47" target="_blank" rel="noopener noreferrer">View on GitHub</a>
Expand All @@ -129,16 +129,16 @@ <h1>AgentGuard vs LangSmith vs Langfuse vs Portkey</h1>
<!-- Key differentiators -->
<div class="highlights">
<div class="highlight-card">
<strong>Runtime intervention</strong>
<strong>Stops agents, not just tracks them</strong>
<p class="muted">AgentGuard kills agents mid-run when they exceed spend limits. Others only report after the damage is done.</p>
</div>
<div class="highlight-card">
<strong>Zero dependencies</strong>
<p class="muted">Pure Python stdlib. One package, nothing to audit. No supply chain risk, no dependency conflicts.</p>
<strong>Hosted policy management</strong>
<p class="muted">Set budget caps, loop limits, and timeouts from the dashboard. No code changes to update limits.</p>
</div>
<div class="highlight-card">
<strong>Free and open source</strong>
<p class="muted">MIT-licensed SDK. No per-trace pricing for local use. Dashboard optional.</p>
<strong>Free OSS + flat-rate hosted</strong>
<p class="muted">MIT-licensed SDK works offline. Hosted dashboard starts free, scales with flat monthly pricing. No per-trace charges.</p>
</div>
</div>

Expand Down Expand Up @@ -178,14 +178,7 @@ <h2>Feature comparison</h2>
<td><span class="no">No</span></td>
</tr>
<tr>
<td>Cost tracking</td>
<td class="col-ag"><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span></td>
</tr>
<tr>
<td>Tracing / spans</td>
<td>Cost tracking per run</td>
<td class="col-ag"><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span></td>
Expand All @@ -199,67 +192,32 @@ <h2>Feature comparison</h2>
<td><span class="partial">Partial</span> &mdash; gateway-level only</td>
</tr>
<tr>
<td>Rate limit guard</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; per-minute throttling</td>
<td><span class="no">No</span></td>
<td><span class="no">No</span></td>
<td><span class="yes">Yes</span> &mdash; gateway-level</td>
</tr>
<tr>
<td>Runtime dependencies</td>
<td class="col-ag"><span class="yes">Zero</span></td>
<td>5+</td>
<td>3+</td>
<td>3+</td>
</tr>
<tr>
<td>Open source SDK</td>
<td class="col-ag"><span class="yes">MIT</span></td>
<td><span class="no">Proprietary</span></td>
<td><span class="yes">MIT</span></td>
<td><span class="partial">Partial</span></td>
</tr>
<tr>
<td>Self-hosted option</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; SDK works fully offline</td>
<td><span class="no">No</span></td>
<td>Spend visibility dashboard</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; per model, per day</td>
<td><span class="yes">Yes</span></td>
<td><span class="no">No</span></td>
</tr>
<tr>
<td>CI cost gates</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; GitHub Action included</td>
<td><span class="no">No</span></td>
<td><span class="no">No</span></td>
<td><span class="no">No</span></td>
</tr>
<tr>
<td>LangChain integration</td>
<td class="col-ag"><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span> &mdash; native</td>
<td><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span></td>
</tr>
<tr>
<td>LangGraph integration</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; guarded_node decorator</td>
<td><span class="yes">Yes</span> &mdash; native</td>
<td><span class="partial">Partial</span></td>
<td>Policy management (hosted)</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; budget caps, loop limits</td>
<td><span class="no">No</span></td>
<td><span class="no">No</span></td>
<td><span class="partial">Partial</span> &mdash; gateway config only</td>
</tr>
<tr>
<td>CrewAI integration</td>
<td class="col-ag"><span class="yes">Yes</span></td>
<td><span class="no">No</span></td>
<td>Open source SDK</td>
<td class="col-ag"><span class="yes">MIT</span></td>
<td><span class="no">Proprietary</span></td>
<td><span class="yes">MIT</span></td>
<td><span class="partial">Partial</span></td>
<td><span class="no">No</span></td>
</tr>
<tr>
<td>OpenAI/Anthropic auto-patch</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; one-line patching</td>
<td><span class="partial">Via wrapper</span></td>
<td><span class="yes">Yes</span></td>
<td><span class="yes">Yes</span> &mdash; gateway proxy</td>
<td>Zero dependencies</td>
<td class="col-ag"><span class="yes">Yes</span> &mdash; Python stdlib only</td>
<td>5+ deps</td>
<td>3+ deps</td>
<td>3+ deps</td>
</tr>
</tbody>
</table>
Expand Down Expand Up @@ -312,28 +270,28 @@ <h2>Pricing comparison</h2>
</div>

<!-- Detail sections -->
<h2>Why AgentGuard for budget enforcement?</h2>
<h2>Why teams choose AgentGuard for cost control</h2>
<div class="detail-grid">
<div class="detail-card">
<h3>LangSmith tracks costs. AgentGuard enforces them.</h3>
<p class="muted">LangSmith shows you what an agent spent after it finishes. AgentGuard raises <code>BudgetExceeded</code> at the dollar limit you set and stops the agent immediately. The difference between a dashboard alert and a circuit breaker.</p>
<p class="muted">LangSmith shows you what an agent spent after it finishes. AgentGuard raises <code>BudgetExceeded</code> at the dollar limit you set and stops the agent immediately. The difference between a report and a circuit breaker.</p>
</div>
<div class="detail-card">
<h3>Langfuse is open source. So is AgentGuard.</h3>
<p class="muted">Both are MIT-licensed. Langfuse focuses on tracing and prompt management. AgentGuard adds runtime guards &mdash; budget limits, loop detection, and timeout enforcement that stop agents before they cause damage.</p>
<h3>Langfuse traces. AgentGuard controls spend.</h3>
<p class="muted">Both are MIT-licensed. Langfuse focuses on tracing and prompt management. AgentGuard focuses on cost control &mdash; budget limits, loop detection, and spend visibility that keep your AI bill predictable.</p>
</div>
<div class="detail-card">
<h3>Portkey is a gateway. AgentGuard is a library.</h3>
<p class="muted">Portkey proxies all LLM traffic through their servers. AgentGuard runs in your process with zero network calls. No latency overhead, no data leaving your infrastructure, no single point of failure.</p>
<h3>Portkey proxies traffic. AgentGuard runs in-process.</h3>
<p class="muted">Portkey routes all LLM calls through their gateway. AgentGuard runs in your process &mdash; no latency, no data leaving your infrastructure, no single point of failure.</p>
</div>
<div class="detail-card">
<h3>Zero dependencies means zero risk.</h3>
<p class="muted">Every dependency is supply chain attack surface. AgentGuard uses Python stdlib only. One package to install, one package to audit. No transitive vulnerabilities, no version conflicts.</p>
<h3>Hosted dashboard, not just an SDK.</h3>
<p class="muted">The SDK enforces limits locally. The dashboard gives your team spend visibility, policy management, and alert feeds &mdash; without building your own infrastructure.</p>
</div>
</div>

<!-- Code example -->
<h2>Add budget enforcement in 3 lines</h2>
<h2>Add cost control in 3 lines</h2>
<p class="muted">No signup required. No API keys. Works offline.</p>
<pre class="code"><span class="kw">from</span> agentguard <span class="kw">import</span> Tracer, BudgetGuard, patch_openai

Expand Down
Loading