Skip to main content

Migrating from Sourcegraph / Cody

Mental-model shift​

Sourcegraph is a hosted, organisation-wide, cross-repo code-intelligence platform. Cody is the AI layer on top. The core value props:

  • One search over every repo in the org.
  • Cloud indexing; no per-developer local footprint.
  • Integrates with code-host permissions.

Shaktiman is local-first, single-project, developer-machine-bound:

  • Indexes one project at a time.
  • Lives on your laptop; no external service.
  • Designed for the MCP ecosystem (Claude Code, Cursor, Zed).

These are different products solving different problems. You'd pick Shaktiman when a single developer needs rich, agent-usable code intelligence for a project they work on day-to-day, and you don't want (or can't have) a hosted service in the loop.

Feature parity table​

CapabilitySourcegraph / CodyShaktimanNotes
Cross-repo search✓✗Single-repo only.
Hosted, zero per-developer install✓✗Shaktiman runs per-dev.
Respects host-level permissions✓N/ARuns on your laptop; your permissions are yours.
Free for open-source reposPartiallyAlwaysMIT.
Ranked semantic search✓✓Similar idea, different implementation.
AI chat over codebaseCodyClaude Code / Cursor / Zed + ShaktimanSubstitute the client.
Precise cross-language references✓PartialShaktiman's graph is per-language.
Offline / air-gapped✗✓Shaktiman has no external dep except optional Ollama.
No data leaves the machine✗✓ (sans Ollama-as-a-service configurations)Important for some orgs.
GraphQL API / integrations✓✗ (MCP instead)Different integration model.

Side-by-side workflow​

"Find every caller of validateToken across our org."

Sourcegraph:

  1. Open Sourcegraph UI or call the API.
  2. validateToken → references panel → spans all repos.
  3. Click through results.

Shaktiman:

  1. You can only do this per repo. If validateToken exists in repo A and callers are in repo B, Shaktiman indexed on repo A doesn't see repo B.
  2. Workaround: run Shaktiman on each repo, query each.

This is a genuine gap, not a workaround that closes it.

Gaps (where Sourcegraph wins)​

  • Cross-repo. Full stop. If that's your need, Shaktiman is the wrong tool.
  • Permissions-aware. Sourcegraph enforces who-can-see-what; Shaktiman runs as you, on your files.
  • Infrastructure for large orgs. Shared indexes, shared tuning, shared observability. Shaktiman has none.
  • Cody as an integrated product. Cody is the Sourcegraph AI layer. With Shaktiman you supply your own client (Claude Code, Cursor, etc.).

Where Shaktiman wins​

  • Local-first. No server to stand up, no data egress, no licenses.
  • Zero cost for open-source or internal use.
  • Fits the MCP ecosystem. If your agent is Claude Code / Cursor / Zed, Shaktiman is native.
  • Budget-fitted context assembly. context is explicitly designed to produce a token-bounded context package — a primitive Sourcegraph doesn't expose directly.

When to keep Sourcegraph / Cody​

  • Your team works across many repos and needs unified search.
  • Your org's security model requires hosted, permission-aware code access.
  • You already have Sourcegraph as an infra decision and it serves you.

You can run both. Sourcegraph for cross-repo exploration; Shaktiman for day-to-day, agent-driven work inside the repo you're currently editing.

See also​