SuperBased
OPEN SOURCE · APACHE-2.0
PRODUCT OVERVIEW

OBSERVER

Universal Observability, Monitoring & Cost Control for AI Coding Agents

See, govern, and cut the cost of every AI coding agent your team runs — Codex, Claude Code, Cursor, Cline, Copilot, and the rest — from one local-first control plane.

▶ PRESS START npm i -g superbased-observer
THE BLIND SPOT

Your team adopted AI agents overnight. The controls didn't come with them.

Spend, quality, and audit are invisible — and each tool only shows you its own slice.

84%
use AI coding tools
Up from 76% a year ago. Adoption is universal and accelerating.
$200-600
real cost / dev / mo
Tools advertise $10–20/mo. Actual loaded cost runs 10–30× higher.
29%
trust AI output
Down from 40% last year. Adoption is outrunning confidence.

No unified view across agents, spend, quality, or sessions. Every vendor hands you a different dashboard — and none of them see the others. The missing layer is the control plane around all of them.

THE CONTROL PLANE

One control plane across every agent. Not another agent.

SuperBased Observer turns scattered agent activity into measurable, governable infrastructure.

WHAT IT DOES
  • Captures token usage and actions — edits, reads, timing, shell commands, API turns — across every coding agent.
  • Normalizes all of it into one shared, project-centric model — instead of one dashboard per vendor.
  • Makes it queryable through a dashboard, CLI, and MCP — for humans and agents alike.
WHY IT MATTERS
  • Spend attribution by model, project, session, and tool.
  • Full audit trail — every agent action observed and recorded, system-wide.
  • Token compression that cuts spend with no loss in quality.
  • Operational context for failures, retries, and repeated work.

From black box to system of record. Observe everything. Normalize it once. Feed the useful parts back into better agent decisions.

THE FEEDBACK LOOP

It doesn't just watch your agents. It helps them work better.

A four-stage loop, rendered like game levels 1 → 2 → 3 → 4.

LV 01
Capture
Logs, hooks, and proxy data become a clean event stream.
→
LV 02
Reason
Freshness, failure context, patterns, and redundant work are inferred.
→
LV 03
Act
MCP tools answer the exact questions agents need before they spend more tokens.
→
LV 04
Improve
Compression, shared memory, and routing hints make future runs better.
↻ FEEDBACK LOOP Past agent behavior becomes input to future agent decisions: file & command freshness, failure context, session recovery, cost rollups.
PRODUCT OVERVIEW

One binary. Seventeen integrations. Two package managers.

Distributed via npm and PyPI — the same version-locked binary. Plus a VS Code extension.

observer — zsh — 80×24
$ npm i -g superbased-observer
+ superbased-observer@1.10.0 · pypi mirror ok
$ observer up
[ok] reverse proxy  :8820
[ok] dashboard      :8081
[ok] mcp server    16 read-only tools
[ok] sqlite store  wal mode
[ok] jsonl adapters 17 integrations
watching agents… █
5-COMPONENT ARCHITECTURE
JSONL adaptersreverse proxy :8820SQLite storedashboard :8081MCP · 16 tools
CROSS-PLATFORM
Linux x64 / arm64macOS Intel + Apple SiliconWindows x64
INTEGRATIONS  (and growing)
CodexClaude CodeCursorCline / RooCopilotCopilot CLIOpenCodeOpenClawPiAntigravity (Gemini)HermesKilo CodeCline CLICowork
THE FULL SURFACE

Five jobs, one install.

Observer isn't a single chart — it's the control layer for the whole agent stack.

OBSERVE
Per model / project / session / tool spend, an action audit trail, live session detail, and a local dashboard. Accurate tokens via the proxy.
COMPRESS
Shell & conversation compression cut token spend ~10% with no quality loss — with per-command savings telemetry.
GOVERN
A security & control layer: observe-then-enforce rules over commands and agent actions, with an audit trail mapped to SOC 2 & NIST 800-53.
ROUTE & FORECAST
Model-routing waste detection (flags Opus where Sonnet would do), cache-efficacy tracking, and a next-message cost & limit predictor.
REMEMBER
16 read-only MCP tools let agents query freshness, failure context, and project knowledge before spending more tokens.
SAVINGS · ROI

Pay for it in the first month — out of the spend it removes.

Compression + routing + cache visibility attack the bill from three sides.

THE LEVERS
  • Compression — context & shell-output compression cut token spend ~10% with no deterioration in output quality.
  • Routing waste — the analysis engine surfaces over-powered sessions (Opus where Sonnet would do) and quantifies the savings before you act.
  • Cache efficacy — cache-hit tracking + a model-switch forecast tell you when prompt caching is, and isn't, earning its discount.
~10%
token spend removed by compression — no quality loss
80K+
tracked turns
On the founders' own daily usage — organic, deep validation.
$20K+
API cost monitored
Dogfooded daily across every adapter.
13.3K
npm downloads
Dual npm + PyPI · zero paid marketing.
LOCAL-FIRST

It runs entirely on your machine. Nothing leaves it.

The control plane that doesn't become a new data-exfiltration risk.

POSTURE
  • No telemetry. No cloud. No data leaves the machine — the default install is 100% local.
  • Secrets scrubbed at the boundary — content and command output are never stored; only paths, commands, excerpts, and metrics.
  • Open source, Apache-2.0 — auditable end to end, no black box.
  • Single self-contained binary — embedded SQLite (WAL), no agents or background services you didn't ask for.
0
bytes leave the machine by default
WHEN YOU GO TEAM
  • Node-side opt-in, never org-forced — hashed metadata by default; raw content only if the node operator enables it.
  • Bring-your-own-cloud — host the Teams server in any cloud. Cloud-agnostic, no hyperscaler lock-in.
TEAMS · ENTERPRISE

Land with one developer. Roll out to the whole org.

When multi-user visibility becomes critical, Observer scales without changing its identity.

WHAT TEAMS ADDS
  • Fleet analytics — spend, usage, and quality across every developer and every agent, in one rollup.
  • Multi-user dashboards — per-team, per-project, per-model views.
  • SSO + audit logs — SAML auth and an audited message-content viewer.
  • Org policy distribution — push signed Guard & model-routing policies to the fleet; nodes apply them locally.
  • Bring-your-own-cloud — runs in any cloud you choose; privacy posture stays node-controlled.
HOW IT STAYS SAFE

Privacy stays node-controlled. The org admin has no remote toggle to force raw content off a developer's machine — by design.

Open-source adoption drives the land; cost visibility and governance pull Observer into the team budget.
WORKS WITH WHAT YOU RUN

No rip-and-replace. Observer wraps the agents you already use.

Vendor-neutral by design — it sits across the stack, so it complements every tool instead of competing with one.

INSTALL IN MINUTES
  • npm i -g superbased-observer (or PyPI, or the VS Code extension) → observer start → done.
  • Auto-detects and registers with every AI tool it finds; opt out of any piece with a flag.
  • Cross-platform — Linux x64/arm64 · macOS Intel + Apple Silicon · Windows x64.
5 MIN
to first dashboard · 17 integrations and counting
PARTNER / CO-SELL ANGLE
  • Neutral infrastructure — for agent vendors, IDE makers, and platform teams, Observer is the measurement & governance layer that makes your tool safe to deploy at scale, without forcing the customer onto a single agent.
  • Adapter framework — new surfaces are added quickly; the roadmap is an adapter on every AI-coding surface.
USE CASES

If your team ships with AI agents, Observer is the missing instrument panel.

Same install, four jobs-to-be-done.

Engineering leaders
“What are we spending on AI coding, by team and project — and is it worth it?”
→ Unified spend attribution + ROI.
Platform / DevEx
“Standardize observability and guardrails across whatever agents people pick.”
→ One control plane, any tool.
FinOps / finance
“Cut the bill without slowing engineers down.”
→ ~10% compression savings + routing-waste detection.
Security / compliance
“Prove what the agents did — and enforce limits.”
→ Guard audit trail mapped to SOC 2 / NIST, local-first.
CONTINUE? ▶ YES

Start free. Scale when you're ready.

Two ways in — adopt the open-source core today, talk to us when you need the fleet.

▸ PLAN 01
Free · Open-source core
$0
Observer · watcher · MCP server (16 tools) · CLI analytics · dashboard · shell + context compression. Apache-2.0. Local-first.
Adopt without procurement. Saves tokens on day one.
▸ PLAN 02
Enterprise · Teams & governance
TALK TO US
Bring-your-own-cloud telemetry · fleet analytics · multi-user dashboards · SSO + audit logs · org policy distribution.
For engineering leadership when multi-user visibility & governance become critical.
npm i -g superbased-observer && observer start docs & demo · superbased.app/observer