Integrations

Connect Praxiom AI with GitHub, Linear, Jira, and Google Drive for end-to-end product execution.

Overview

Praxiom AI integrates with four platforms so the agent can ground recommendations in your real work and push decisions out to where execution lives. Each integration follows the same shape — OAuth connect, scoped resource selection (repos, team, or project), agent tools, and a sync path that flips Praxiom state when the external ticket closes.

This page is an index. Each platform has a dedicated guide with its full tool catalog, endpoint list, and error semantics.

Pick Your Platform

PlatformAgent toolsWrite tools (HITL)Primary use
GitHub7 read, 1 writecreate_github_issueGround recommendations in real code; push issues with status sync back to implemented
Linear5 read, 3 write (+1 issue creator)create_linear_issue, update_linear_issue, create_linear_commentQuery backlogs, cycles, projects; update state/assignee/cycle under HITL
Google Drive5 read, 3 writegoogle_drive_write, google_drive_update, google_drive_batch_importSearch, read, write files; bulk-import Drive folders as research sources
JiraNone— (execution-only)Terminal stage of the execution-ticket pipeline with P0–P3 priority mapping

Common Patterns

All four integrations share:

  • OAuth state lives in a DB row with a 10-minute expiry and is deleted on read (one-time use).
  • Tokens are encrypted at rest with the workspace-level key. If decryption or the external API returns 401, the integration surfaces a typed TOKEN_ERROR and the user reconnects.
  • Feature gating — GitHub uses has_github; Linear, Jira, and Google Drive use has_integrations.
  • Integration model — one row per (workspace_id, integration_type) holding the encrypted token, optional refresh token, expiry, is_active flag, connected_at, last_synced_at, last_error, and a JSON config dict with platform-specific metadata (selected repos / team / cloud_id / Google profile).
  • Manual sync — GitHub and Linear both expose POST /sync; closed/completed issues flip their linked recommendation to status="implemented".

HITL Approval

Every write tool across integrations is decorated with @gated(preview_fn=...). Calling a gated tool from chat returns { status: "queued", pending_action_id } — the user approves the action from the pending-actions inbox before it executes. The approval preview renders a pure, deterministic summary of exactly what will be written (issue fields, Drive content preview with char/word count, Linear diff).

See Pending Actions for the full approval flow.

Execution Tickets

Beyond per-tool pushes, the execution pipeline generates tickets from document blocks and pushes them in bulk to any of GitHub, Linear, or Jira — with priority mapping:

PraxiomGitHubLinearJira
P0(labels)1 (urgent)Highest
P1(labels)2 (high)High
P2(labels)3 (medium)Medium
P3(labels)4 (low)Low

See Execution Tickets for the generate → push → sync workflow.

What's Next

Was this helpful?