GUIDE

Cursor, VS Code, Zed, Claude Code

Use Taskclan Intelligence inside your editor. Taskclan exposes two drop-in compatibility surfaces — OpenAI and Anthropic — so any client that already speaks either wire format works with a Taskclan API key. All calls debit the same T1 credit wallet.

Endpoints

Get a key from platform.taskclan.com/api-keys.

Model names

Four model ids, one for each tier plus Auto. Pick Auto to let Taskclan decide per prompt (recommended for most editor use); pick a specific tier when you want deterministic cost. Unrecognized names fall back to Flow.

Capabilities

Every tier is a full first-class model in your editor — tools, vision, PDFs, prompt caching, and streaming all work end-to-end. Extended thinking (Cursor Plan / Fable Plan) is Max-only, and Auto promotes prompts with reasoning cues to Max automatically.

Cursor

Cursor renders /v1/models entries verbatim in its model picker — the goal is to see T1-auto, T1-core, T1-flow, T1-maxappear alongside "GPT-5.6", "Sonnet 5", "Grok 4.5", etc. Cursor's own built-in models keep using your Cursor subscription, so the two live side-by-side.

1. Open Cursor Settings (⌘, on macOS, Ctrl+, on Windows/Linux).
2. Click the Models tab in the left sidebar of the settings panel.
3. Scroll to Custom API Keys → OpenAI API Key. Toggle Override OpenAI Base URL ON.
4. Fill in the two fields (get your key from platform.taskclan.com/api-keys):

   Base URL: https://engine.taskclan.com/api/openai/v1
   API Key:  sk_t1_...

5. Click Verify. Cursor probes GET /v1/models — a green check means Taskclan responded. A red X almost always means the base URL or key is wrong.
6. In Model Names (same panel), add the four ids if they aren't auto-populated: T1-auto, T1-core, T1-flow, T1-max. Click Save.
7. Open any Cursor chat or composer (⌘L / ⌘I). Click the model dropdown at the bottom of the input. Scroll — you should see T1-auto, T1-core, T1-flow, T1-max in the list.
8. Pick T1-auto and send a message. If a reply comes back, you're on Taskclan — the router picked Core / Flow / Max for you based on the prompt shape, and every turn debits your T1 credit wallet.
9. Enable Agent mode (⌘L → toggle Agent). T1 tiers advertise full agent capabilities via /v1/models — the "Agent may not work well" pill should not appear. Pick a task like "read this file and propose a change" to confirm the tool loop round-trips.

VS Code — Taskclan Intelligence extension

The primary path for VS Code users. Our native extension adds @taskclan to VS Code's built-in chat, with /agent, /plan, /refactor, and /explain slash commands. Auto-routes across T1 Core / Flow / Max per prompt. Same wallet, no config file, first-class brand.

1. Install from the Marketplace — search "Taskclan Intelligence" in the Extensions view, or run:

   code --install-extension taskclan.taskclan-intelligence

2. Command palette → Taskclan: Set API Key → paste your sk_t1_... key from platform.taskclan.com/api-keys.
3. Open Chat (⌘⌥I / Ctrl+Alt+I) → type @taskclan hi. The reply streams back attributed to Taskclan Intelligence, with a footer showing which T1 tier ran, the credit cost, and your remaining balance.

Slash commands pick the tier for you:

Works in Cursor too — Cursor is a VS Code fork, so the extension installs and runs there identically. Cursor Chat doesn't consume the built-in chat-participant API, though, so if you want @taskclan inside Cursor Chat specifically, configure Cursor's built-in custom OpenAI provider instead — see the Cursor section above. The extension still gives Cursor users the participant surface in Cursor's VS Code-style Chat panel.

VS Code — Continue.dev (alternative)

If you already use Continue.dev and prefer keeping everything in one extension, T1 tiers work as OpenAI-compatible models via Continue's config. This is the older path — most VS Code users should just install the Taskclan Intelligence extension above. Continue's model picker reads from ~/.continue/config.yaml. Configure once, then all four tiers show up as distinct entries in the Continue sidebar picker.

1. Install the Continue extension from the VS Code Marketplace if you haven't already (Marketplace ID: Continue.continue).
2. Open the Continue sidebar (⌘L or the Continue icon), then click the gear → Open Config File. That opens ~/.continue/config.yaml.
3. Add one block per tier under models:. Replace sk_t1_... with a key from platform.taskclan.com/api-keys.

models:
  - name: T1 Auto
    provider: openai
    model: t1-auto
    apiBase: https://engine.taskclan.com/api/openai/v1
    apiKey: sk_t1_...
    roles: [chat, edit, apply]
    capabilities: [tool_use, image_input]
  - name: T1 Core
    provider: openai
    model: t1-core
    apiBase: https://engine.taskclan.com/api/openai/v1
    apiKey: sk_t1_...
    roles: [chat, autocomplete]
    capabilities: [tool_use, image_input]
  - name: T1 Flow
    provider: openai
    model: t1-flow
    apiBase: https://engine.taskclan.com/api/openai/v1
    apiKey: sk_t1_...
    roles: [chat, edit, apply]
    capabilities: [tool_use, image_input]
  - name: T1 Max
    provider: openai
    model: t1-max
    apiBase: https://engine.taskclan.com/api/openai/v1
    apiKey: sk_t1_...
    roles: [chat, edit, apply]
    capabilities: [tool_use, image_input]

Two things about that config are worth knowing:

• provider: openai names Continue's internal transport adapter — it doesn't mean your requests go to OpenAI. Continue's provider registry is a fixed allowlist (openai / anthropic / gemini / …) and rejects unknown names at config load, so OpenAI-compatible services (Groq, Together, Fireworks, Taskclan) all reuse provider: openai with a custom apiBase. The actual endpoint is https://engine.taskclan.com/api/openai/v1.
• The capabilities: [tool_use, image_input] line enables tools and image input across Chat, Plan, and Agent modes. Continue's mode picker may still render a small ⚠️ triangle next to Plan/Agent for T1 tiers — that's driven by a separate hardcoded name-regex in Continue's source (isRecommendedAgentModel), is cosmetic, and doesn't affect functionality. If it bothers you, use model: claude-sonnet-4-5-t1-flow (or claude-haiku-4-5-t1-core, claude-opus-4-5-t1-max, claude-sonnet-4-5-t1-auto) — the server resolver accepts any string ending in -t1-{tier}, so the prefix is opaque payload we use to satisfy Continue's allowlist.

4. Save the file. Continue picks up changes on save — no reload needed for most versions. If it doesn't: ⌘⇧P → Continue: Reload Config.
5. Click the model name at the bottom of the Continue sidebar chat. You should see T1 Auto, T1 Core, T1 Flow, T1 Max in the list — none with the warning triangle. Switch to Agent or Plan mode and send a message; the reply comes back via T1 and debits your wallet.

Zed

Zed speaks the Anthropic wire format natively — point its anthropic provider at our shim and the T1 tiers appear in Zed's Assistant model picker.

1. Open Zed's settings.json (⌘, → gear → Open Settings JSON, or~/.config/zed/settings.json).
2. Add a language_models.anthropic block that points at Taskclan and declares the three tiers:

"language_models": {
  "anthropic": {
    "api_url": "https://engine.taskclan.com/api/anthropic/v1",
    "available_models": [
      { "name": "T1-auto", "display_name": "T1 Auto", "max_tokens": 200000 },
      { "name": "T1-core", "display_name": "T1 Core", "max_tokens": 200000 },
      { "name": "T1-flow", "display_name": "T1 Flow", "max_tokens": 200000 },
      { "name": "T1-max",  "display_name": "T1 Max",  "max_tokens": 200000 }
    ]
  }
}

3. Save. Open the Assistant panel (⌘? or the assistant icon).
4. Command palette → Assistant: Set Anthropic API Key. Paste your sk_t1_… key.
5. Click the model dropdown at the top of the Assistant panel — T1 Auto / Core / Flow / Maxshould be listed under Anthropic. Pick T1 Auto and send a message — the router picks a tier per prompt.

Claude Code (CLI)

Anthropic's Claude CodeCLI reads two environment variables to swap providers wholesale. Set both, run claude, and every turn hits Taskclan Intelligence.

1. Install Claude Code if you don't have it (npm install -g @anthropic-ai/claude-code).
2. Export the two variables (persist them in ~/.zshrc or ~/.bashrc so every new terminal picks them up):

export ANTHROPIC_BASE_URL="https://engine.taskclan.com/api/anthropic"
export ANTHROPIC_API_KEY="sk_t1_..."

# Optional: pin the tier claude uses. T1-auto is usually the right pick —
# the router chooses per prompt. Otherwise: T1-core / T1-flow / T1-max.
export ANTHROPIC_MODEL="T1-auto"       # recommended

3. Reload your shell (source ~/.zshrc or open a new terminal).
4. Run claude "Reply with exactly: PONG". If you see PONG come back, you're on Taskclan.
5. Every subsequent claude invocation — interactive or one-shot — routes through T1 and debits your wallet.

Claude Desktop (MCP)

Claude Desktop can't swap its underlying model, but it can invoke external tools via the Model Context Protocol. The @taskclan/mcp package exposes Taskclan Intelligence as four tools — taskclan_run(any T1 tier), taskclan_studio_generate (image), taskclan_workflow_run(named engine workflow), taskclan_balance (wallet). Once configured, Claude can decide to call them mid-conversation.

Open ~/Library/Application Support/Claude/claude_desktop_config.json(create it if it doesn't exist) and add one entry under mcpServers:

{
  "mcpServers": {
    "taskclan": {
      "command": "npx",
      "args": ["-y", "@taskclan/mcp"],
      "env": {
        "TASKCLAN_API_KEY": "sk_t1_..."
      }
    }
  }
}

Restart Claude Desktop. Open a new chat, click the tools icon in the composer — you should see taskclan_run, taskclan_studio_generate, taskclan_workflow_run, taskclan_balance listed under a Taskclan section. Ask Claude "use Taskclan Max to compare X and Y" and it'll call the tool; the response streams back into the chat and debits your T1 wallet.

The same block works in Claude Code (~/.claude/settings.json) and in Zed's settings.json under an mcp key — any MCP-capable client picks it up.

VS Code — Cline

Cline (formerly Claude Dev) is a VS Code agent that speaks the Anthropic wire format. Point it at Taskclan and the T1 tiers become picker options in the Cline sidebar.

1. Install Cline from the VS Code Marketplace (Marketplace ID: saoudrizwan.claude-dev).
2. Open the Cline sidebar (its icon in the activity bar), then click the gear (Settings).
3. API Provider → select Anthropic Compatible.
4. Fill in the three fields:

API Base:  https://engine.taskclan.com/api/anthropic/v1
API Key:   sk_t1_...
Model:     T1-auto    (or T1-core / T1-flow / T1-max)

5. Click Done. Send a message in the Cline chat — the reply routes through T1.
6. To pin a specific tier, change the Model field to T1-core, T1-flow, or T1-max. Cline reloads the provider on save.

aider (CLI)

aider is a terminal AI pair-programmer. Point it at Taskclan via two env vars and it uses the T1 tier you pass on the command line.

1. Install aider (pip install aider-chat or see aider.chat).
2. Export the two environment variables — persist them in ~/.zshrc or ~/.bashrc:

export OPENAI_API_BASE=https://engine.taskclan.com/api/openai/v1
export OPENAI_API_KEY=sk_t1_...

3. Run aider with an explicit T1 tier:

   aider --model openai/T1-auto                    # recommended — router picks per prompt
   aider --model openai/T1-flow                    # or pin a specific tier
   aider --model openai/T1-core file.ts             # start a session on file.ts

4. The openai/ prefix tells aider to use its OpenAI provider (which reads the env vars above). Every turn hits Taskclan and debits your wallet.

curl

Both endpoints are plain HTTP — great for smoke-testing before you wire an editor:

# OpenAI shape — T1-auto lets the router pick per prompt (see x_taskclan.autoRoutedFrom / autoRoutedReason).
curl https://engine.taskclan.com/api/openai/v1/chat/completions \
  -H "Authorization: Bearer sk_t1_..." \
  -H "Content-Type: application/json" \
  -d '{"model":"T1-auto","messages":[{"role":"user","content":"Reply PONG"}]}'

# Anthropic shape — same idea; x_taskclan sidecar reports the resolved tier.
curl https://engine.taskclan.com/api/anthropic/v1/messages \
  -H "x-api-key: sk_t1_..." \
  -H "Content-Type: application/json" \
  -d '{"model":"T1-auto","max_tokens":16,"messages":[{"role":"user","content":"Reply PONG"}]}'

Billing and errors

Both endpoints meter against the same wallet as /api/t1/complete and the SDK. A 402 means insufficient credits — see Errors & rate limits for the recommended catch pattern. All other error codes match the upstream wire format (invalid_request_error, authentication_error, etc.).