Skip to main content
If you don’t find the answer you’re looking for here, reach out to our support team at support@agenttherapy.ai — we’re happy to help.

​
Getting Started

Agent Therapy works with any LLM-powered agent. We provide first-class support for:
  • OpenAI Assistants — direct thread import via the OpenAI API
  • LangChain — native callback handler for automatic log capture
  • AutoGen — built-in message listener integration
If you’re using a custom agent framework, you can submit conversation logs through our REST API. As long as your agent produces structured conversation data, Agent Therapy can analyze it.
No. Agent Therapy requires zero code changes to your agent. Simply export your agent’s conversation logs in any of our supported formats and submit them for analysis. If you’d like deeper, real-time integration (such as automatic session triggering), optional SDK callbacks are available — but they’re never required to get started.
Most users run their first diagnostic session within 5 minutes of signing up. The typical flow is:
  1. Create your account and grab your API key from the dashboard.
  2. Export a conversation log from your agent.
  3. Submit the log via the dashboard or API.
  4. Review your Agent Health Score and findings.
No lengthy onboarding, no sales call required.

​
How It Works

Agent Therapy analyzes your agent’s conversation logs across four dimensions:
  • Conversation patterns — identifies repetition, loops, and off-topic drift
  • Tool call sequences — detects redundant, missing, or out-of-order tool usage
  • Output consistency — flags contradictions and unstable responses across similar prompts
  • Instruction adherence — measures how closely the agent follows its system prompt
Detection combines rule-based heuristics (fast, deterministic checks) with ML models trained on thousands of real-world agent failure modes. Findings are ranked by severity so you know what to fix first.
The Agent Health Score is a single 0–100 metric that summarizes your agent’s behavioral quality for a given session.The score is computed from a weighted combination of all findings in a session. Individual finding categories (consistency, tool use, instruction adherence, etc.) are also broken out so you can pinpoint problem areas quickly.
Drill into the Findings tab of any session to see exactly which behaviors contributed to the score and get recommended fixes.
For production agents, we recommend triggering a session after every significant conversation batch — for example, at the end of each day or after every 100 conversations. This lets you catch regressions early before they affect end users.You can automate this entirely via the API so sessions run on a schedule or after every deployment. For non-production or experimental agents, running on-demand is perfectly fine.
Yes. Agent Therapy doesn’t require a live connection to your agent. You can submit conversation logs at any time — even weeks after they were recorded. This makes it ideal for:
  • Post-incident analysis
  • Auditing historical agent behavior
  • Comparing behavior across model versions
Simply upload your logs via the dashboard or the /sessions API endpoint and analysis will begin immediately.

​
Data & Privacy

Yes. All data is protected at every layer:
  • In transit — encrypted with TLS 1.3
  • At rest — encrypted with AES-256
Your conversation data is never used to train our models or shared with third parties. Each customer’s data is logically isolated within our infrastructure.
Enterprise customers can request a full security and compliance report including our SOC 2 Type II attestation.
By default, session logs and findings are retained for 90 days. You can adjust this — including reducing it to as few as 7 days — in your account under Settings → Data Retention.You can also delete individual sessions or all data for a specific agent at any time from the dashboard or via the API.
Data is stored in US data centers (AWS us-east-1) by default. If you need data residency in the European Union, the EU region (AWS eu-west-1) is available on the Enterprise plan.Contact sales@agenttherapy.ai to discuss regional data requirements before signing up.

​
Plans & Pricing

Agent Therapy offers three plans:All plans include the full feature set. Enterprise adds SSO, dedicated support, EU data residency, and custom data retention policies.Visit our pricing page for current rates and a detailed feature comparison.
If you hit your monthly session limit, you have two options:
  • Purchase add-on sessions — available in blocks from the billing page; they activate immediately and carry over for 30 days.
  • Let them queue — sessions submitted over the limit are queued and processed automatically once your limit resets at the start of your next billing cycle, in the order they were received.
Queued sessions are not processed until your limit resets or you purchase add-ons. If timely analysis is critical, upgrade your plan or buy add-ons before you hit the ceiling.

​
Integrations

Yes. Agent Therapy is fully model-agnostic — it analyzes conversation structure and behavior, not model internals. It works with any LLM, including:
  • OpenAI GPT-4o, GPT-4, GPT-3.5
  • Anthropic Claude 3 and newer
  • Google Gemini
  • Meta Llama and open-source models
  • Any fine-tuned or self-hosted model
As long as you can export a conversation log, Agent Therapy can analyze it.
Yes. Agent Therapy exposes webhooks and a full REST API so you can push findings into whatever observability tools you already use. Native integrations include:
  • Datadog — pipe health scores and finding counts into custom metrics and dashboards
  • PagerDuty — trigger incidents automatically when an agent’s health score drops below your threshold
  • Slack — post session summaries and critical alerts to any channel
Configure integrations in the dashboard under Settings → Integrations, or wire up custom workflows using the webhook payload directly.