ADLC Try now
← Back to Blog
Setup

Integrating ADLC with Your Tools

8 min read • Aug 11, 2026

ADLC is designed to work with the tools your team already uses. This guide walks through configuring repository hosting, issue tracking, and model providers.

Repository Hosting

ADLC supports both GitHub and GitLab. The integration allows agents to clone, branch, commit, and create pull/merge requests.

GitHub

  • Setup: Create a GitHub App with the "Contents", "Pull Requests", and "Checks" permissions, or use a personal access token with repo and workflow scopes.
  • Security best practice: Use fine-grained personal access tokens scoped to specific repositories where possible. Rotate monthly.
  • Deployment: Store the token in the organization vault (encrypted per-org or use a managed key if deployed by the platform).

GitLab

  • Setup: Create a project access token or group access token with api, read_repository, and write_repository scopes.
  • Security best practice: Bind tokens to specific projects when possible. Set expiration dates and rotate periodically.
  • Deployment: Same vault pattern as GitHub.

Issue Tracking

ADLC agents read and update tickets from your tracking system. Supported platforms:

Linear

  • Setup: Generate an API key from Linear settings. The key allows reading and updating issues, creating sub-issues, and managing relationships.
  • Best practice: Create a dedicated "AI Fleet" bot user in Linear, then generate the API key for that bot. Makes audit trails clear.

Jira

  • Setup: Create a service account with "Jira Software" and "Jira Service Management" access, then generate an API token.
  • Note: Requires Jira Cloud (jira.atlassian.net). On-premise Jira Server is not supported.

Asana

  • Setup: Create a personal access token scoped to the projects where agents will work.
  • Limitation: Asana's API does not support creating parent-child task relationships programmatically, so agents cannot auto-decompose tasks into subtasks. Plan decomposition must be manual or use an intermediate representation.

Model Providers

ADLC agents can use different LLM providers for planning, implementation, and testing. Configure these in your workspace settings.

Anthropic (Claude)

  • API key: Generate from console.anthropic.com. Store in the organization vault.
  • Model selection: Use Claude 3.5 Sonnet for planning and review, Opus for research-heavy tasks, Haiku for lightweight per-token tasks.

OpenAI (GPT)

  • API key: Create from platform.openai.com. Requires payment method.
  • Model selection: Use GPT-4 for complex tasks, GPT-4 Turbo for faster execution, GPT-3.5 Turbo for cost optimization on simple tasks.

Local Models (Ollama, LM Studio)

  • Setup: Point ADLC to a local model endpoint (e.g., http://localhost:11434 for Ollama).
  • Best for: Experimentation, privacy-sensitive work (medical, legal), or air-gapped environments.
  • Tradeoff: Slower inference than cloud providers. Requires powerful hardware to run larger models.

Gemini (Google)

  • API key: Generate from makersuite.google.com.
  • Note: Gemini is cost-effective for high-volume tasks but has lower latency tolerance (some endpoints have global rate limits).

Configuration Patterns

Here are common configurations by team size and use case:

Startup (GitHub + Linear + Claude)

  • GitHub for repositories (one org, multiple repos).
  • Linear for planning (easier to use than Jira at small scale).
  • Claude for planning and coding (best reasoning, reliable).

Enterprise (GitHub + Linear + Multiple Providers)

  • GitHub for primary development, GitLab for infrastructure-as-code.
  • Linear for product work, Jira for ops/infra tracking.
  • Claude for complex planning, GPT-4 for coding (lower cost at scale), Ollama for sensitive work.

Privacy-First (Self-Hosted)

  • GitLab self-hosted for code.
  • Jira self-hosted for tracking.
  • Ollama local for all inference (zero external API calls).

Multi-Provider Strategies

Some teams use multiple providers for different lifecycle stages:

  • Purpose routing: Planning uses Claude (best for reasoning), coding uses GPT-4 (fast and reliable), testing uses Haiku (cheap per-token).
  • Vendor diversity: Fallback to a secondary provider if the primary is rate-limited or down (e.g., Claude → OpenAI for failover).
  • Cost optimization: Use Haiku for fast tasks under 5 minutes, Claude for complex tasks over 30 minutes, GPT-4 for balanced cost/performance.

Troubleshooting Integration Issues

Common problems and solutions:

  • "Unauthorized" when accessing repositories: Check that the repository access token has the correct scopes (read + write). Generate a new token if needed. For GitHub Apps, verify the installation is enabled for the org.
  • Agents cannot find or update tickets: Verify the issue tracking token has access to the target projects. For Jira, confirm the account has the right project permissions, not just workspace permissions.
  • LLM rate limiting: If agents are being rate-limited, reduce concurrency (run fewer parallel agents) or distribute load across multiple API keys.
  • Stale credentials: If a token has been rotated, manually update it in the vault. ADLC does not auto-detect stale credentials; it fails on the first request.

Read the full brief: ADLC Brief
Back to blog: All posts