Built on Claude.
CodeTractor is a Claude-native platform. We use Anthropic's Claude model family — Haiku, Sonnet, and Opus — to power every agent decision, with intelligent routing that picks the right model for each task.
The Claude Model Family
Three models, one routing layer. The right brain for each task.
Claude Haiku 4.5
Triage & Classification
Classifies incoming tasks from Slack, Teams, or CLI. Routes to the right specialist agent in under half a second.
- Intent detection
- Ticket triage
- Log parsing
- Structured extraction
Claude Sonnet 4.6
Code & Execution
Writes production code, opens PRs, runs tests, and drives deployments. The workhorse behind most agent actions.
- Code generation
- Refactoring
- Test authoring
- Tool use
Claude Opus 4.7
Reasoning & Root Cause
Deep reasoning for incident diagnosis, architecture decisions, and multi-step debugging. The model we reach for when stakes are highest.
- Root cause analysis
- Architecture review
- Incident diagnosis
- Migration planning
Haiku classifies → Sonnet executes → Opus reasons when stuck
Every Claude capability, wired in.
We don't just call the API — we use the full Claude platform.
1M Context Window
Opus 4.7 with 1M tokens means whole repos, weeks of logs, and full incident history fit in one prompt — no chunking, no retrieval hacks.
Extended Thinking
Claude's thinking blocks let agents reason through multi-step problems transparently. You see the full reasoning chain in Mission Control.
Tool Use & MCP
Every agent is built on Anthropic's tool-use primitives and Model Context Protocol. Clean, auditable, and model-agnostic wiring for every integration.
Prompt Caching
We cache your codebase context, SKILL.md files, and runbooks. 90% cost reduction on repeat work — the Deploy Agent doesn't re-read the same Terraform file ten times a day.
Sub-Agents (Agent SDK)
Built on the Claude Agent SDK. Supervisor dispatches parallel specialists, each with their own context window, reducing cross-contamination and cost.
Responsible Scaling
Anthropic's Responsible Scaling Policy, constitutional AI safety training, and zero data retention on API traffic — trust you can inherit.
CodeTractor vs. raw Claude API
Claude is the engine. CodeTractor is the platform around it.
| Feature | Raw Claude API | CodeTractor |
|---|---|---|
| Claude model family | You choose one model per call | Auto-routes Haiku → Sonnet → Opus |
| Multi-agent orchestration | You build it | 5 pre-built specialists, Supervisor included |
| Human-in-the-loop gates | You build it | PR approval, deploy gates, Slack/Teams built in |
| Episodic memory | Stateless | Agents learn from past incidents across sessions |
| SKILL.md capability discovery | N/A | Drop-in runbooks, no redeployment |
| Auto-rollback on failure | N/A | DevOps Agent reverts and alerts automatically |
| Cost optimization | Manual | 60–70% lower via routing + caching |
| Deployment target | Anthropic API | Your VPC, Bedrock, or Claude Managed Agents |
How your data is handled
- Your code and logs never train shared models.
- Anthropic does not train on API traffic by default.
- Self-hosted deployments keep all data in your VPC.
- SOC 2 Type II aligned, zero retention on inference.
Claude is the engine. Let us build the platform around it.
Free assessment. We'll map your engineering workflows to Claude-powered agents in a 45-minute call.