Free integration demo

Test Trajbl in your own AI app in minutes.

Generate a free pilot key, copy one setup prompt into your coding agent, and let the agent verify the API, safety rules, and token savings before it changes your system.

Trajbl is a hosted compression layer for agent and RAG systems. Your app keeps its own LLM provider, routing, billing, and approval flow.

Select your app type

Choose Captain Claw or a chat app backend such as OpenAI, Gemini, Ollama, OpenRouter, or LiteLLM.

Generate your free API key

The key is shown once. Store it only in a local or server-side environment file.

Copy the setup prompt

Your coding agent appends safe env placeholders, asks you to paste the key, and runs only synthetic smoke tests first.

Agent gives green light

The agent reports whether the API, key, compression, fail-open behavior, and metrics are safe before permanent integration.

Step 1

Select your app.

Choose the integration target. The selected module controls the generated key category and the setup prompt.

Available pilot

Chat apps

OpenAI, Gemini, Ollama, OpenRouter, LiteLLM, and similar chat-message backends.

Use adaptive compression for support/RAG context before the provider call. The technical format is OpenAI-style chat messages, but the final model provider can be different.

Available pilot

Captain Claw

The kstevica/captain-claw coding-agent project.

Add a fail-open Trajbl wrapper before Captain Claw calls its downstream LLM provider.

View Captain Claw on GitHub
Chat apps is selected.Now generate your API key, then use the matching integration prompt below.
Step 2

Generate access.

No email server is used. The key is shown once on this page. Copy it immediately and store it only in server-side environment variables.

Pilot keys are intended for testing with synthetic or non-sensitive data. Usage limits may apply.
API key will be created for: Chat apps
Step 3

Copy the agent prompt.

After key generation, copy this prompt into your coding agent. It will append env placeholders, ask you to paste the key locally, run synthetic smoke tests, and report if integration is safe.

Chat apps
Generate your API key first, then copy this module-specific prompt.
You are my coding agent. Safely prepare a Trajbl integration for this chat app backend.

Goal:
Add Trajbl as an optional, fail-open context-compression step immediately before the app sends chat messages to its LLM provider.

This module works with apps that use OpenAI-style chat message arrays, including OpenAI, Gemini adapters, Ollama OpenAI-compatible endpoints, OpenRouter, LiteLLM, vLLM, LM Studio, LocalAI, and similar backends.

Strict safety rules:
- Do not call a real LLM provider during setup.
- Do not send private app data to Trajbl during the first smoke test.
- Do not print, log, commit, screenshot, or hardcode the Trajbl API key.
- Keep the API key in a server-side/local environment file only.
- Keep the integration OFF by default until I explicitly approve it.
- If Trajbl fails, times out, returns invalid JSON, or changes protected hard context, pass the original messages through unchanged.

Step 1: Inspect read-only first.
Find the narrowest point where final chat messages are sent to the provider. Do not edit files yet.

Expected integration shape:
app builds messages
-> optional Trajbl wrapper
-> POST https://api.neurobiz.me/v1/compress
-> provider receives compressed messages
-> app receives provider response

Step 2: Prepare environment configuration.
Find the existing .env or local environment config. If it exists, append missing variables without overwriting existing values. If it does not exist, propose the safest local env file.

Add this placeholder, then tell me exactly where to paste the API key value I received from the Trajbl page:
TRAJBL_PROXY_ENABLED=0
TRAJBL_API_URL=https://api.neurobiz.me
TRAJBL_API_KEY=PASTE_TRAJBL_KEY_HERE
TRAJBL_COMPRESSOR=stable
TRAJBL_COMPRESSION_MODE=adaptive
TRAJBL_TIMEOUT_MS=3000
TRAJBL_FAIL_OPEN=1
TRAJBL_LOG_CONTENT=0

Step 2b: Prepare local metrics.
Create a local project folder named .trajbl if it does not exist.
Configure the wrapper to append one JSON line per Trajbl attempt to:
.trajbl/metrics.jsonl

The metrics file must not contain prompt text, user messages, retrieved documents, API keys, or provider responses.
Only write numeric/status fields such as timestamp, enabled, raw_input_tokens, compressed_input_tokens, token_savings_percent, compression_latency_ms, fallback_used, fallback_reason, compression_mode, hard_context_preserved, client_id if available, and provider name if safe.

Step 3: Verify safely.
After I paste the API key locally, run only safe checks:
- GET https://api.neurobiz.me/health
- GET https://api.neurobiz.me/v1/usage with the Bearer key
- POST https://api.neurobiz.me/v1/compress with a tiny synthetic dummy payload
- use a fake downstream provider for any OFF vs ON smoke test

Step 4: Report before implementation.
Give me a short report:
- API reachable: yes/no
- Key valid: yes/no
- Compression works on dummy data: yes/no
- Savings metric visible: yes/no
- Local metrics path created: yes/no, exact path
- No secrets exposed: yes/no
- Proposed files to change
- Rollback plan
- Exact command or setting to turn Trajbl ON later

Do not implement permanent integration until I write exactly:
APPROVE TRAJBL INTEGRATION
Available module

Captain Claw integration target

Captain Claw is a coding-agent project. Trajbl's pilot module is designed to sit before Captain Claw's downstream LLM provider call, compressing long context while keeping the integration optional and fail-open.

Open kstevica/captain-claw

What Trajbl preserves

  • system prompts
  • current user request
  • code blocks
  • JSON/config
  • tool calls and results
  • stack traces and logs

What Trajbl compresses

  • older conversation history
  • long notes
  • documentation-like prose
  • retrieved support/RAG context
  • repeated background context

How to read results

  • fallback_used=false means Trajbl compressed successfully.
  • fallback_used=true means original messages were passed through safely.
  • token_savings_percent measures input-context reduction before the provider call.

Learn the full flow

Read the documentation for API routes, metrics, safety rules, troubleshooting, and expected pilot behavior.

Open documentation